Download Accurate Quantification Of Image Blurring On Thin Bone Thickness

Accurate Quantification Of Image Blurring On Thin Bone Thickness And Material Properties By Micro-CT
1,2
Maloul,A ,1,2Fialkov, J, +1,2Whyne, C
+1University of Toronto, Toronto, ON, 2 Sunnybrook Health Sciences Centre, Toronto, ON
Senior author [email protected]
INTRODUCTION:
For more than three decades computed tomography (CT) has been
widely used for diagnostic and quantitative assessment of bone in
patients. Improvement of the CT resolution and computation power led
more recently to the utilization of CT images in the development of 3D
finite element (FE) models to predict failure loads and fracture patterns
for bone structures. Patient specific FE models of complex thin bone
structures can be generated through CT imaging via direct conversion
into FE meshes. However, developing such models is limited by
geometric complexity and imaging techniques. The accuracy of thin
bone FE models largely depends on accurate representation of specimen
geometry and material properties. Characterizing bone thickness in very
thin bone structures is difficult because its thickness is smaller than the
resolution of diagnostic CT imaging system. The size of any structure is
limited to the minimum size of the radiation beam size. In addition, bone
structures smaller than the beam size become blurred proportionally to
the size of the beam. Low-resolution scans due to large beam size
resulting in blurred images have been shown to yield inaccurate
representation of geometry and material properties of thin bone
structures1. However, previous studies have not provided a Micro-CT
based quantification of the overestimation in the thickness or
underestimation of the intensity1, 2, nor have they provided information
about the effect of overestimation and underestimation on the strain
patterns in thin bone structures. Previous work in our lab showed that
stress and strain patterns in thin bone FE models exhibit high sensitivity
to changes in bone thickness and material properties3.
The objectives of this work were to: 1. Quantify the impact of scan
resolution on the thickness and intensity profile of thin bone structures
and cortical thickness measurements. 2. Assess the effect of resolution
on strain patterns in thin bone structures through 3D finite element
modeling. 3. Investigate the effect of upsampling and smoothing
algorithms on decreasing image blurring. We hypothesize that clinical
CT scans resolutions result in large overestimations of the thickness of
thin bones and concurrent decreases in image intensity profiles due to
image blurring resulting in an inaccurate representation of strain patterns
in thin bone FE models.
METHODS:
µCT scans of a human acetabulum and sinus bone were acquired
from a preserved pelvis and craniofacial skeleton at a resolution of
4.1µm and 13.75 µm respectively (SkyScan 1172 and GE Healthcare
eXplore Locus). Both scans were down sampled to 16.4 µm, 82 µm, 164
µm, 328 µm, and 488 µm to evaluate the effect of image resolution on
bone thickness and intensity profile (Fig 1). 16.4 µm was used as the
highest resolution in this study as no changes in bone thickness or
intensity were observed between scans at 4.1 µm and 16.4 µm. The
scans at different resolutions were segmented to identify the boundaries
of the bone using an intensity based threshold criteria (AmiraDEV3.1).
The segmentations were used to generate 3D surfaces and automatically
construct tetrahedral FE meshes at resolutions from 16.4 µm (569339
elements) to 488 µm (7101 elements). Bone thickness and intensity
profiles were quantified at each resolution using full 3D surfaces. The
3D FE meshes were used to solve for maximum principal strain values
under a 100N axial compressive load (Abaqus 6.7-1). The acetabulum
and sinus scans at 488 µm were then upsampled to 16.4 µm and
smoothed using smoothing algorithms. The upsampled scans were
segmented and used to create 3D surfaces and FE models to evaluate the
bone thickness and maximum principal strain.
RESULTS:
The minimum measured cortical thickness in the acetabulum was
1.221 mm at a resolution of 16.4µm. The minimum measured thickness
in the sinus was 0.564 mm at 16.4µm. At a 488µm resolution blurring
resulted in an increase in measured bone thickness of 71.9% and 116.5%
in the acetabulum and sinus respectively (Fig 2). A 26.39% and 10.87%
decrease were observed in the maximum scan intensity as resolution
decreased from 16.4µm to 488µm in the acetabulum and sinus scans
respectively. Maximum principal strain decreased by 71.85% in the
acetabulum as resolution decreased from 16.4µm to 488µm. Similarly,
maximum principal strain decreased by 51% in the sinus as resolution
decreased to 488µm. Low resolution scans resulted in regions of concen-
trated strain in the FE models due to non smooth edges. A 95%
averaging threshold was used to eliminate edge effects causing high
strains at these regions yielding comparable maximum strains to
smoothed models at an equivalent resolution. Upsampling and
smoothing of the low resolution scans resulted in 53.72% decrease
in the bone thickness overestimation in the acetabulum while a 32.39%
decrease in the bone thickness overestimation was observed in the sinus
bone. Maximum principal strain in the FE models developed using the
upsampled scans were similar to the maximum principal strain in the FE
models developed from the high resolution scans (16.4µm). Upsampling
the scans did not correct for the underestimation of the scan intensity.
Figure 1: Sinus scan at 16.4 µm (right) adjacent to the same scan
at resolution of 488 µm (left)
Figure 2: Bone thickness measurements at resolutions from 16.4
to 488 µm
DISCUSSION:
Insufficient CT resolution can result in blurring and jagged edges and
represents a major limitation in developing accurate FE models of thin
bone structures. Increases in bone thickness measurements and
decreases in maximum image intensity due to image blurring
significantly affected the FE generated strain patterns. Smoothing was
able to eliminate edge effects in low resolution images. Maximum strain
did not vary substantially with changes in resolution ranging from 164
µm - 488 µm. In contrast, large changes in maximum strain were found
to occur as resolution increased from 164 µm - 16.4 µm. Currently it is
not possible to use micro CT scanning for diagnostic purposes.
However; upsampling diagnostic CT scans and smoothing can decrease
the overestimation in bone thickness estimates due to image blurring by
over 30%. This will allow for the development of more accurate FE
models to investigate the biomechanical behavior of thin bone structures
and will enable more accurate quantification of cortical bone thickness
for assessing bone quality in osteoporotic patients. This work
demonstrates that high-resolution imaging is required to accurately
represent thin bone geometry and material properties within thin bone
structures. Modeling of mechanical behaviour in thin bone regions using
FE analysis will be limited if direct conversions from clinical resolution
CT scans are utilized.
REFERENCES:
1-Louis O. Investigative radiology. 1993 Sep;28(9):802-5
2-Hangartner TN et al. Physics in medicine and biology,1987
Nov;32(11):1393-406
3- Szwedowski T, Thesis, 2007 Oct : P 115-116
Poster No. 749 • 55th Annual Meeting of the Orthopaedic Research Society