Medical Image Processing - Signal Processing Systems - page 367

Fig. 15 Example of surface rendering ( left ) and volume rendering ( right )

set of voxels with the same intensity and techniques like marching cubes convert
a discrete set of voxels to a continuous surface. Multiple thresholds may be used
to construct several isosurfaces. To match with existing GPU paradigms, surfaces
tend to be modeled as polygons, which then go through the traditional graphics
processing pipeline of texture/color selection, shading and rasterizing. An example
of surface rendering is shown in Fig. 15 .

8.2 Volume Rendering

It is not necessary to identify the voxels of interest and make the associated
assumptions in surface rendering to visualize in 3D space an object of interest.
An alternative to surface rendering, volume rendering is a more brute force albeit
computationally demanding approach to 3D visualization. In volume rendering, all
voxels are assigned to one or more tissue types based on their intensity. Fuzzy
memberships are allowed such that a particular voxel may belong to two classes
(e.g., 60% muscle and 40% fat). Such classification techniques are less error prone
and eliminate the need for single threshold-based isosurfacing techniques in surface
rendering. Associated with each tissue type is also a color and an opacity (or
translucency). These can be controlled to hide or highlight different intensity ranges
or tissue types. In fact, most visualization systems offer predefined color and opacity
functions (called, presets) to achieve this effect.
After color and opacity assignments and voxel-wise shading calculation, volume
rendering is the process of creating a composite projection image from a 3D set
of voxels. Ray tracing techniques are common and the primary computational chal-
lenge is to resample the 3D image along each ray at a fine sampling interval and then