Direct Volume Deformation
Florian Schulze, Katja Bu¨hler, and Markus Hadwiger
VRVis Research Center, 1220 Vienna, Austria
{fschulze,buehler,msh}@vrvis.at
http://medvis.vrvis.at
Abstract. This paper presents an integrated approach for interactive direct vol-
ume deformation and simultaneous visualization. The fundamental requirement
is that interactive performance without pre-processing must be achieved for large
volume data, where at any time up to one million elements participate in a defor-
mation that is applied interactively by picking and dragging in the 3D view. Cur-
rent physically-based approaches are still one or two orders of magnitude away
from this goal. In contrast, our approach extends the non-physical ChainMail
algorithm and combines it with on-the-fly resampling and GPU ray-casting. Spe-
cial transfer functions assign material properties depending on volume density.
The affected subvolume is deformed and resampled onto a rectilinear grid on the
CPU, and updates the volume on the GPU where it is rendered using ray-casting.
While the deformation is already being displayed, its quality is simultaneously
refined via an iterative relaxation procedure executed in a parallel thread.
Keywords: Deformation, Resampling, Volumerendering.
1 Introduction
This paper follows a vision first published in 1995: Thought as natural extension to
direct volume rendering, Sarah F. Gibson formulated the idea for a system that al-
lows direct deformation, cutting and carving of volume data [6]. She introduced the
so called ChainMail algorithm allowing in its extension modeling of deformation of in-
homogeneous materials. Similar to direct volume rendering, the deformation is directly
performed at the voxel level of the volume without any pre-processing.
The ChainMail algorithm provides only a non physics based deformation scheme,
but is able to deform large structures in real time: Having in mind that a small vol-
ume dataset of 2563 consists already of more than 16 million voxels, existing phys-
ically based approaches are still far away from being able to deform such structures
at interactive frame rates without previous simplification. Due to the limited available
computational power at the time of first publication of the algorithm and its extensions,
simultaneous volume rendering of the whole dataset during the deformation process
was not possible, and Sarah Gibson formulated this task as future work [8].
This paper presents a framework that integrates high quality real time visualization
with direct deformation of volume data fulfilling the following requirements:
– Full information of the original data is available throughout the whole process: De-
formation and visualization are directly performed at the voxel level of the volume.
J. Braz et al. (Eds.): VISIGRAPP 2007, CCIS 21, pp. 59–72, 2008.
c© Springer-Verlag Berlin Heidelberg 2008

60 F. Schulze, K. Bu¨hler, and M. Hadwiger
– The deformation is not physically correct, but plausible depending on the underly-
ing data.
– No time-consuming preprocessing is necessary, like segmentation, simplifications
and adaptive hierarchy generation.
– The system reaches interactive frame rates for simultaneous simulation and visual-
ization.
Basis of the proposed deformation system is the Enhanced ChainMail algorithm [16]
that is taken as initialization step for a relaxation solver that allows also simulation of
elastic deformation. Handling of the high amounts of data has been addressed by a
specialized data structure and memory management system. A new image order re-
sampling algorithm has been developed to provide simultaneous visualization of the
deformed data using the powerful GPU accelerated volume rendering framework de-
scribed in [17].
The paper is organized as follows: Related work is discussed in the next section. A
short summary of the Chain Mail algorithm and existing extensions is given in section
3. Section 4 outlines the general workflow of our system. The two-step deformation
method is explained in section 5 including details on the basic chain mail implementa-
tion, relaxation, and material definition. Visualization and related issues are addressed
in section 6, interaction methods are discussed in section 7. The paper closes with re-
sults in section 8, and a summary in section 9.
2 Related Work
Detailed discussion of the extensively available related work on physically based defor-
mation methods of (volumetric) objects is beyond the scope of this paper. The interested
reader is referred to two State of the Art Reports presented at Eurographics 2005 [14,3]
giving an excellent general overview.
Considering physically based approaches for direct deformation and visualization of
volume data, modern point based mesh free methods [12] seem to be the most natural
approach to deal directly with medical volume data: theoretically, no preprocessing
is required and deformation could be directly performed on the volume if each voxel
would be modeled as particle or phyxel.
The approaches mentioned above and reported in [14,3] provide physically correct
deformation, but due to their computational complexity, none of them is able to handle
more than 100k elements at interactive frame rates, even if GPU accelerated integration
schemes are used [9,11]. Simultaneous visualization of deformed objects is another bot-
tleneck, especially if surfaces have to be reconstructed on the fly, like it is the case in
general for particle-based and point-based approaches [1]. Nealen et al. [14] stated in
the conclusions of the state of the art report: ”Yet even with the current methodology,
the algorithms and models have seen somewhat limited application in production envi-
ronments and videos games. One reason for this is the lack of computational power...”.
Existing approaches addressing directly the deformation of volumes, i.e. without pre-
vious mesh extraction and/or simplification, are mainly based on space or ray deforma-
tion techniques: either a coarser structure (e.g. bounding boxes [18], volume or surface
geometry [23]) is deformed and the deformation of the volume itself is computed as