Fast Volume Segmentation With Simultaneous Visualization
Using Programmable Graphics Hardware
Anthony Sherbondy Mike Houston
Stanford University
Sandy Napel∗
(a) (b) (c) (d)
Figure 1: These four volume renderings utilize a fully opaque transfer function, but are segmented using the method discussed in this paper.
The segmented volumes show: (a) abdominal aortic branch vessels, (b) an aortic aneurysm, (c) an aorta, and (d) peripheral blood vessels in
the lung. The yellow arrows indicate the location of the user’s initial seeds that were evolved to form the presented segmentations.
Abstract
Segmentation of structures from measured volume data, such as
anatomy in medical imaging, is a challenging data-dependent task.
In this paper, we present a segmentation method that leverages the
parallel processing capabilities of modern programmable graphics
hardware in order to run significantly faster than previous methods.
In addition, collocating the algorithm computation with the visual-
ization on the graphics hardware circumvents the need to transfer
data across the system bus, allowing for faster visualization and in-
teraction. This algorithm is unique in that it utilizes sophisticated
graphics hardware functionality (i.e., floating point precision, ren-
der to texture, computational masking, and fragment programs) to
enable fast segmentation and interactive visualization.
CR Categories: I.4.6 [Segmentation]: Region Growing,
Edge and Feature Detection; I.3.7 [Computer Graphics]: Three-
Dimensional Graphics and Realism—Volume Rendering; I.3.8
[Computer Graphics]: Applications
Keywords: region growing, diffusion, segmentation, graphics pro-
cessor, streaming computation
∗
{sherbond,mhouston,snapel}@stanford.edu
1 Introduction
In current radiological practice, highly trained medical imaging
specialists segment anatomical regions of interest from computed
tomography (CT) and magnetic resonance (MR) volumes for fur-
ther analysis. The segmentation process involves the trained oper-
ator selecting the voxels that belong to the anatomy of interest by
drawing contours around cross-sectional views and linking these
cross-sections together.
This manual procedure can be an extremely tedious chore be-
cause of the complexities of anatomical structures. For instance,
arterial vessels may take many odd shapes as they twist and turn
throughout the anatomy making them very difficult to track along
the multitude of cross-sections of the volume. The vessels may
also not have clearly defined edges separating them from nearby
objects of similar intensity (i.e., bone for contrast enhanced images
and soft tissue otherwise) and may have multiple regions where the
anatomy stops and then begins again due to constrictions or scan-
ner artifacts. Manually segmented data is impacted by the user’s
training and judgement. For example, one must decide whether or
not to include calcium deposits, thrombus, constricted regions, and
different portions of the vessel.
Many contributions have been made to the field of automatic seg-
mentation. However, the complicated structures found in medical
imaging offer several unsolved challenges to automated algorithms,
including the lack of global defining morphological characteristics,
scanner noise and artifacts, and an incomplete or weak separation
between voxels representing neighboring tissue. Any of the exis-
tent automated algorithms can be shown to fail on certain datasets
for reasons specific to each algorithm [Kirbas and Quek 2003; Lev-
enton et al. 2000].
Another major drawback of the automated algorithms to date is
that they offer limited “trial and error” based user interaction. The
user often sets global parameters and runs the algorithm hoping to
get the correct results. If the desired result is not achieved, the
user attempts to adjust the parameters and runs the algorithm again.
171
IEEE Visualization 2003,
October 19-24, 2003, Seattle, Washington, USA
0-7803-8120-3/03/$17.00 ©2003 IEEE
Proceedings of the 14th IEEE Visualization Conference (VIS’03)
0-7695-2030-8/03 $ 17.00 © 2003 IEEE