Interactive GPU-based Segmentation of Large Medical Volume Data with Level-Sets

Automatic or semi-automatic segmentation of anatomical structures in three-dimensional datasets is still an ambitious challenge in medical image computation. The scope of application contains visualization, diagnosis, and preplanning of operations. This paper presents a 3D GPUbased level-set solver for segmentation which focuses on segmenting and rendering large volumes. Neither the volume nor the level-set have to be resident on the GPU in their entirety. Only those parts of the volume which are relevant for computation are held in memory and are dynamically swapped in and out of GPU memory. The levelset solver is integrated in a high-quality GPU based ray caster.