The Ultrasound Visualization Pipeline - Scientific Visualization: Uncertainty, Multifield, Biomedical, and Scalable Visualization - page 273

Fig. 24.3 Automatic segmentation of the ovarian follicles [ 13 ]
the system can narrow down the search for other features. On the other hand, segmen-
tation is an extremely critical procedure which may obscure diagnostically relevant
aspects of the anatomy under examination. Consequently, fully automatic segmen-
tation techniques have not been implemented in clinical systems so far, with the
exception of a method for follicle volumetry [ 13 ], as shown in Fig. 24.3 .
A great challenge with ultrasound segmentation is that the data is dependent on
many factors. For one, different positions and orientations of the probe, while looking
at the same anatomical part, can provide very different images. Hyper-echoic regions
cast shadows onto the tissue behind it according to the probe position. This alone,
makes ultrasound segmentation data highly uncertain. Most segmentation techniques
return a model with no indication of the uncertainty of the result. To compensate for
the fuzzy nature of the ultrasound data, Petersch et al. developed a soft segmentation
technique for 3D ultrasound data [ 56 ]. This technique calculates a probability map
for 3D ultrasound data, which in turn can be used to create soft representations of
the features extracted.
24.4.1 Clipping
Feature extraction can be computationally costly. In-vivo 3D ultrasound examination
cannot always afford the extra time necessary to extract the interesting structures.
Therefore, clipping is a commonly used tool in live visualization of 3D ultrasound.
Interactively removing regions which are not interesting, the user gets a clear view
of the features normally occluded. Sakas et al. developed a clipping tool in their
ultrasound visualization system [ 64 ] which is nowadays a standard feature in com-
mercial 3D ultrasound systems. The user can segment the dataset in-vivo using three
approaches: drawing on one of the three axially-aligned slices, selecting everything
along the current axis, and within the sketch. Another tool is based on sketching