Volume visualization of object interiors
Ingmar Bitter, Arie E. Kaufman, Klaus D. Mueller · 2002
We present work on volume visualization of object interiors. Given sampled data with a segmented region of interest, the task is to efficiently and meaningfully display the density field in the interior of this region. In modern medicine, doctors are faced with this problem during each of their virtual endoscopy procedures, such as virtual bronchoscopy, virtual colonoscopy or virtual examination of blood vessels. Our work provides solutions to this problem that include automatic navigation within the region of interest and that create volume rendered views of the sampled density volume along the way. It is imperative that the views created during a virtual endoscopy are rendered as perspective projections of diagnostic quality. The perspective projection is needed to properly simulate the view of a physical endoscope. We analyze the image quality dependencies on interpolation, classification, sampling rate, and compositing buffer depth. Based on this analysis, we present algorithms and architectures to generate perspective views. The algorithms focus on generating high quality perspective projection images on general-purpose hardware with the optimal number of samples. Our volume rendering architecture supports perspective projection through an incremental slice sweep algorithm. Implemented as special-purpose hardware, it can route and process volumetric data much more efficiently than general-purpose computers and can process multiple data streams in parallel. Therefore, it can achieve much higher frame rates. For complete insight into an objects' interior all of the inner surface has to be visualized through several volume rendered views. This is enabled through our guided navigation paradigm that can smoothly transition between automatic and manual navigation of the virtual camera. For manual navigation we use a distance from boundary field to avoid collisions with the object wall. For automatic navigation we provide distance field based algorithms to compute an object's centerline and/or skeleton and then navigate the camera along the centerline or skeleton. Additionally, the automatic navigation can start anywhere inside the object, not only on the centerline or skeleton. The results of this work have been integrated into a virtual colonoscopy system that many radiologists use daily to facilitate diagnosing colon polyps in thousands of patients.