Direct volume rendering methods compute images of a volumetric data set without explicit extraction of geometric surfaces from the data. Optical model is used to map data values to optical properties; namely, color and opacity. Then, the optical properties are projected using a suitable projection function along each viewing ray to render the data. In practical systems, volume data are stored either as a stack of 2D texture slices or as a single 3D texture object. Such data represent samples on a given sampling grid. Values in between grid elements can be computed by interpolating data at neighboring such elements. This process is known as volume reconstruction and has usually demanding computational requirements.