Imaging transforms for visualizing surfaces and volumes

Abstract

Three-dimensional (3D) visualization in biomedical and other imaging areas is a rapidly emerging discipline. The major developments in this field are described in a unified and concise way. To this end, we introduce an operator notation to describe the basic imaging transforms commonly used in 3D visualization and to identify a comprehensive set of basic transforms. We also introduce several new basic transforms for filtering and interpolating scenes and structures and for rendering surfaces and volumes. We demonstrate not only how the existing visualization methodologies can be described concisely, but we also show how a great variety of new methodologies can be generated using both the existing imaging transforms and the new transforms introduced in this paper. A comprehensive evaluation method to compare objectively rendering methods used in visualization based on task-specific mathematical phantoms is described. We examine in detail separate transform sequences that are best suited for rendering robust and frail structures (ie, structures with well- and poorly defined boundaries).

Order-of-magnitude faster isosurface rendering in software on a PC than using dedicated general-purpose rendering hardware

“…For example, the orbits in Figure 4 (top row) exhibit an unnatural brightness and have a "steely" appearance. However, digital surfaces can be smoothed (digitally) using a variety of approaches [30] so that the resulting (smoother) digital surfaces may be rendered at about the same speed as demonstrated in this paper. Both methods do capture a great deal of detail as evidenced by the sutures in Figure 5.…”

Three-dimensional in vivo motion of adult hind foot bones

“…3D reconstruction of each bone was performed by interpolation, 3D Gaussian filtering and thresholding (Raya and Udupa, 1990;Udupa and Goncalves, 1993;Udupa et al, 1991). The resultant binary data consisted of reconstructed virtual images of each of the five bones, anatomically arranged for each imaged foot position.…”

<title>Go digital, go fuzzy</title>

“…These global concepts can be arrived at using simple local concepts for digital surfaces [21,22]. The gradient may also be estimated from other derived scenes such as a Gaussian smoothed version of the segmented binary scene [24]. That is, a face with a normal vector pointing from inside of the surface to its outside in the -x direction is distinguished from a face at the same location with a face normal in exactly the opposite direction.…”