Fast Correction Method for Abdominal Multi-Organ Segmentation Using 2D/3D Free Form Deformation and Posterior Shape Models

Despite advances in automatic medical image segmentation, time-effective interactive correction tools are commonly required in the clinical environment. In this work we present a new interactive correction method for abdominal multi-organ segmentation. Our method enables 3D shape corrections through 2D interactions. During runtime local corrections are used to build a conditional model, which infers the shape of neighboring organs. The combination of 2D/3D shape deformation model with a mechanism to propagate local corrections enables an intuitive and natural manipulation of 3D segmentation results, and reduces substantially the number of interactions and correction time used for multi-organ correction. Experimental results show the ability of the proposed approach to produce high quality segmentations with 82% and 57% reductions on the number of user interactions and correction time, respectively, in comparison to corrections performed on a single-organ basis.