The Ultrasound Visualization Pipeline - Scientific Visualization: Uncertainty, Multifield, Biomedical, and Scalable Visualization - page 277

smooth surfaces and remove unimportant noise in direct volume rendering, Lim
et al. proposed a filtering technique in their GPU based ultrasound rendering frame-
work [ 43 ]. This technique employs different sized filters to smooth out the noise.

24.6.1 Transfer Function Design

For direct volume rendering, transfer functions map ultrasound data, i.e., voxel
echogenicity in B-mode imaging and frequency information in Doppler imaging,
onto colors and opacities. Usually, this mapping is based on look-up tables. In color
Doppler imaging the commonly used red-to-blue color transfer function encodes di-
rection and velocity of flow, whereas a variety of predefined color maps is in use for
B-mode volume rendering. Custom color map editors are available, but hardly ever
used. Overall, there is a well-established set of color-maps used in clinical practice.
Different from color transfer functions, where the selection largely depends on the
preferences of the sonographer, the proper design of an appropriate opacity transfer
function (OTF) is crucial: When designing OTFs, the goal is to assign a high opacity
to voxels of structures of interest, while mapping all other samples to low opacities,
thus avoiding any occlusion of the target structure. Whereas computed tomography
allows classification of tissue based on voxel intensities, tissue classification-based
transfer functions do not work inB-mode imaging due to the completely different data
characteristics. Generally, a high signal intensity arises at a transition between tissues
of different acoustic properties. Thus, at least in the case of soft tissue structures, we
will measure high signal intensity at transitional areas and lower intensity signals
within homogeneous tissue. This is the reason for applying monotonically increasing
opacity transfer functions in DVR of ultrasound data. The aim is to opacify the tissue
transitions in the hope of obtaining a visualization of an entire target structure.
Themost commonly usedOTF in volume rendering of B-mode data assigns voxels
to one of three classes depending on their echogenicity, namely invisible, transparent,
and opaque. The corresponding piecewise linear OTF is modified manually by means
of two parameters, namely a threshold intensity I thresh and a transparency value
α controlling the increase of opacity for intensities above I thresh . The effect of

modifying I thresh is depicted visually on MPR images, see Fig. 24.6 .
The parameters of the OTF affect the rendered image in a substantial way. The
lower the I thresh value, the lower the rendered image's brightness, due to an increas-
ing number of hypoechoic voxels contributing to the image. Furthermore, the OTF
affects depth contrast, i.e., the contrast arising froma spatial discontinuity in the target
structure, and tissue contrast, i.e., contrast due to different echogenicity of adjacent
tissue. See Ref. [ 28 ] for an evaluation of these effects on linear and parabolic OTFs.
On the other hand, any modification of fundamental acquisition parameters, such as,
overall gain, or depth gain compensation, and any change of the position of the trans-
ducer or the target structure, changes the echogenicity distribution and thus requires
modifying the OTF for an optimal visualization. For a real time imaging modality,
incessant modification is not feasible. Hence, in clinical practice sonographers use