Two-way interactive refinement of segmented medical volumes

For complex medical image segmentation tasks which also require high accuracy, prior information must usually be generated in order to initialize and confine the action of the computational tools. This can be obtained by task oriented specialization layers operating within automatic segmentation techniques or by advanced exploitation of user- data interaction, in this case the segmentation technique can conserve generality and results can be inherently validated by the user itself, in the measure he is allowed to effectively steer the process towards the desired result. In this paper we present a highly accurate and still general morphological 3D segmentation system where rapid convergence to the desired result is guaranteed by a two-way interactive segmentation-refinement loop, where the flow of prior information is inverted (from computing tools to the user) in the refinement phase in order to help the user to quickly select most effective refinement strategies.