Section: New Results

Multi-atlas Segmentation

Paticipants: Stavros Alchatzidis, Aristeidis Sotiras, Nikos Paragios

In [33] a novel approach that couples pair-wise deformable registration with multi-atlas segmentation using graphical models was proposed. The method exploits both spaces and seeks to determine the optimal solution which will create the best possible visual agreement between atlases and target image along with their label consistency. The approach optimizes the deformation models and the segmentation labels jointly through an interconnected graph allowing either to relax registration constraints when segmentation labels do indicate or the opposite. The joint optimization of both spaces allowing the “implicit” automatic selection of atlases and therefore improves significantly segmentation performance.