Fig. 1.

Overview of MIST. The initial mesh (dotted line, top left panel) is only roughly aligned to the anatomical structure (dark gray area). At each vertex of this mesh, a perpendicular intensity profile is measured (dotted profile, middle panel) and aligned to the mean profile learned in the training stage (blue profile, middle panel) to find the displacement that aligns the vertex with the anatomy. The displacements of neighbouring vertices are coupled through the MRF shape prior (lower left panel) and combining the profile likelihood (middle panel) with this prior yields the final segmentation (right panel). Full details of the intensity model are given in Visser et al. (2016). Note that the segmentation method is multimodal and multiple intensity profiles are measured at each vertex; this was omitted from the figure for clarity.