Memory efficient acceleration structures and techniques for CPU-based volume raycasting of large data

Most CPU-based volume raycasting approaches achieve high performance by advanced memory layouts, space subdivision, and excessive pre-computing. Such approaches typically need an enormous amount of memory. They are limited to sizes which do not satisfy the medical data used in daily clinical routine. We present a new volume raycasting approach based on image-ordered raycasting with object-ordered processing, which is able to perform high-quality rendering of very large medical data in real-time on commodity computers. For large medical data such as computed tomographic (CT) angiography run-offs (512 /spl times/ 512 /spl times/ 1202) we achieve rendering times up to 2.5 fps on a commodity notebook. We achieve this by introducing a memory efficient acceleration technique for on-the-fly gradient estimation and a memory efficient hybrid removal and skipping technique of transparent regions. We employ quantized binary histograms, granular resolution octrees, and a cell invisibility cache. These acceleration structures require just a small extra storage of approximately 10%.