Fast and Smooth Interactive Segmentation of Medical Images Using Variational Interpolation

Abstract

We present a fast and interactive segmentation method for medical images that allows a smooth reconstruction of an object's surface from a set of user drawn, three-dimensional, planar contours that can be arbitrarily oriented. Our algorithm uses an interpolation based on variational implicit functions. Because variational interpolation is computationally expensive, we show how to speed up the algorithm to achieve an interactive calculation time while preserving the overall segmentation quality. The performance improvements are based on a quality preserving reduction of the number of contour points and a fast voxelization strategy for the resulting implicit function. A huge speedup is achieved by the parallelization of the algorithms, utilizing modern 64-bit multi-core CPUs. Finally, we discuss how to make the interpolation algorithm more robust to selfintersecting and reduced contours.