Depending on the compositing method in use, on the content of look-up tables, and on the rendering mode which defines the usage of the tables, a significant percentage of an object's voxels may not contribute to a rendered image at all. For example, when MIP is used for compositing, black voxels (after the transfer function mapping) do not contribute to the result. If DVR is used, (almost) totally transparent voxels provide no visible contribution. RTVR utilizes a background thread, which is activated whenever the application is idle, to identify those voxels and to reorder data in a way which allows to simply skip non-contributing parts from rendering. The currently visible RenderList is scanned periodically for RenderListEntrys which have not been optimized yet (or which have changed since the last optimization and thus may have to be optimized again). Depending on the rendering parameters of the RenderListEntry a classification of the voxels is performed. Voxels which have been identified as non-contributing are moved to the end of the sequence of clipped respective non-clipped voxels, two pointers are set to indicate the end of voxels which actually have to be rendered (see figure 6.3). For some rendering modes, like the rendering of contours only, opacity and color of voxels change for every new viewing position. In this case the content of the shading LUT is not considered as a criterion for optimization.