The effect of CT slice spacing on the geometry of 3D models

Abstract

THE EFFECT OF CT SLICE SPACING ON THE GEOMETRY OF 3D
MODELS
Beat Schmutz*, Martin E. Wullschleger*, Michael A. Schuetz*^
*Queensland University of Technology, Brisbane, Australia
^Princess Alexandra Hospital, Brisbane, Australia
INTRODUCTION
CT imaging is the current gold standard
for the acquisition of morphological data
for the subsequent 3D reconstruction of
virtual bone models. In addition to
clinical diagnostics, CT image-based 3D
reconstructions are frequently being used
in a broad range of other fields such as
kinematic studies (Dennis et al., 2005),
finite element modelling (FEM) (Penna
et al., 2006; Taddei et al., 2006) and
implant design (Robertson et al., 2000).
Accurate reconstruction of 3D bone
models from CT requires the acquisition
of thin and overlapping slices (Hopper et
al., 1996; Shin et al., 2002). Such high
resolution image data can be obtained
from cadaver specimens without any
problems.
However, in an effort to reduce radiation
exposure to the patient thin sections are
not always clinically justified. Depending
on the application, such data may not
always have sufficient image resolution
to be useful for research purposes.
The current study quantitatively
investigates the effect of various image
slice spacing’s on the geometry of 3D
models of long bones.
METHOD
The femur of a fresh intact ovine cadaver
hind limb was CT scanned at 120KVP
and spacing’s of 0.5, 1, 2, 3 and 4 mm for
the axial slices. The pixel size was 0.4 x
0.4 mm. The image data was
reconstructed using a middle-hard bone
kernel and 50% slice overlap.
3D models of the outer contour of the
femur were generated using the image
processing and registration software
Amira 4.1 (Mercury Computer Systems,
France). The image data was segmented
using a semi-automatic threshold based
method.
The differences in surface geometry
between a model of interest and the
reference model (0.5 mm spaced CT
data) were quantified with the reverse
engineering software RapidForm2006
(INUS Technology). Values were
generated and recorded for the mean
distance, maximum distance and standard
deviation. The percentage fit between a
model of interest and the reference model
was evaluated within a tolerance of ± 0.5
mm.
In addition to quantifying the differences
between the entire surfaces of two
models their deviations for five
anatomical regions (Figure 1) of the bone
were also quantified.
Figure 1. 3D model with anatomical sections:
Distal epiphysis, distal metaphysis, shaft,
trochanter region and femur head.
RESULTS
The effects of the various image slice
spacing’s on the geometry of subsequent
3D models are summarized in Table 1.
The reference 3D model has been
validated against data obtained from a 3D
laser scan (Roland, LPX-250) of the
dissected bone.
CONCLUSIONS
For all models the geometric accuracy
decreased with an increase in the slice
spacing. As expected, the shaft region
was the least affected by variations in
slice spacing’s. The epiphyseal regions
were the most affected due to larger
changes in their surface geometry relative
to the scan axis.
For a tolerance range of ± 0.5 mm, the
results indicate that a slice spacing of 1
mm or less is required to accurately
reconstruct the articulating surfaces of a
long bone. A 2 mm spacing will enable
reconstruction of the metaphyseal areas
with reasonable accuracy. Whereas a
spacing of 3 – 4 mm is sufficient for a
reconstruction of the shaft region.
This demonstrates that the usefulness of
clinically acquired CT data will depend
on the accuracy requirements of the
application for which the 3D models are
intended for.
The findings of this study may also be
useful in the development of optimal
scanning protocols for the imaging of
long bones (Zannoni et al. 1998).
To extend the validation and applicability
of the results of this pilot study, further
tests that involve larger samples and
different bones will be required.
REFERENCES
Dennis et al. J. Biomech. 38:241-253, 2005.
Pena et al. J. Biomech. 39(9), 1686-1701, 2006.
Taddei et al. J. Biomech. 39(13), 2457-2467,
2006.
Robertson et al. J. Bone Joint Surg. 82-A(11),
1594-1602, 2000.
Hopper et al. J. Comput. Assist. Tomo. 20(5),
841-847, 1996.
Shin et al. Korean J. Radiol. 3(1), 49-56, 2002.
Zannoni et al. IEEE Trans Med Imag 17(4), 663-
666, 1998.
Table 1. Differences in 3D model surface
geometries relative to the reference model (0.5
mm spaced CT data) for various CT image slice
spacing’s.
Model
Slice spacing
(mm)
Ave distance
(mm)
Max distance
(mm)
SD
Within ±
0.5mm (%)
Entire femur 1 0.15 0.97 0.12 98
Entire femur 2 0.28 3.08 0.38 86
Entire femur 3 0.40 4.88 0.50 75
Entire femur 4 0.77 6.83 0.83 52
Dist epiphysis 1 0.22 0.91 0.15 96
Dist epiphysis 2 0.44 3.08 0.52 74
Dist epiphysis 3 0.55 4.72 0.55 61
Dist epiphysis 4 0.95 6.29 0.94 42
Dist meta 1 0.12 0.79 0.07 100
Dist meta 2 0.21 1.86 0.26 93
Dist meta 3 0.26 2.46 0.24 87
Dist meta 4 0.65 3.55 0.60 54
Shaft 1 0.10 0.32 0.05 100
Shaft 2 0.09 0.49 0.07 100
Shaft 3 0.12 0.88 0.11 99
Shaft 4 0.26 1.23 0.18 92
Trochanter 1 0.14 0.85 0.10 99
Trochanter 2 0.21 1.67 0.22 93
Trochanter 3 0.43 2.18 0.67 77
Trochanter 4 0.77 2.92 0.83 49
Femur head 1 0.20 0.97 0.16 96
Femur head 2 0.52 2.36 0.48 63
Femur head 3 0.67 4.88 0.58 51
Femur head 4 1.31 6.83 1.04 27
Dr Beat Schmutz
Institute of Health & Biomedical Innovation
Queensland University of Technology
60 Musk Avenue
Kelvin Grove QLD 4059, Australia
Fax: +61 7 3138 6030
Email: b.schmutz@qut.edu.au