3. Curation


Results of two-step curation: (a) The resulting clean model of the patient heart (green) along with the extraneous components (red) which are discarded. Note, the left and right sub-parts are still connected. (b) A stricter merge parameter ( = 5, described in the text) separates the two subparts as desired. (c,d) The left and right parts are shown separately for visual clarity.

We pose the problem of curation as follows: Given a triangulated input surface , possibly with more than one connected component, how can one identify the noisy and irrelevant features and remove them?
The stable manifold of each maximum is a volumetric subset of R3 and is approximated by a collection of Delaunay tetrahedra. The stable manifold of the maximum inherits the same inside/outside marking as that of the maximum itself. For curation, we choose only the inner maxima and compute their stable manifolds. The neighboring stable manifolds are then merged carefully if they satisfy a specific threshold based selection criteria. For our purpose, we initially select a high value of the threshold and merge the adjacent stable manifolds to form large clusters hoping that the patient heart will still survive while the thin blood vessels and barely connected bones and tissues will get detached from it. In fact, as is shown in figure above, that is indeed the case. The thin vessels and the loosely connected bones and other tissues are collected as union of tetrahedra which are simply removed by changing their in-out tag from inside to outside. As we further decrease the parameter , we can separate the left and right part of the patient heart. We must note that ideally they  should not have been connected in the first place, because the inner wall of the left and right part are separated by a muscle wall.