Abstract Archives of the RSNA, 2003
K19-1028
Accurate Nodule Volume Estimation from Helical CT Images: Effect of Reconstruction Filter, Slice Thickness, and Volume Estimation Method
Scientific Papers
Presented on December 3, 2003
Presented as part of K19: Physics (Image Processing: CAD V--Lung)
Metin Gurcan PhD, PRESENTER: Nothing to Disclose
Abstract:
HTML
Purpose: To compare the effects of reconstruction filters, slice thickness
and the volume estimation method on the accuracy of nodule volume estimation
from helical computed tomography (CT) images.
Methods and Materials: A CAD tool for automated assessment of lung nodules
(QuickLookTM, CADx Systems, Beavercreek, OH) was used to segment
nodules from thoracic helical CT images and report characteristics such as
nodule size, volume and average intensity. Nineteen approximately spherical
nodules constructed from modeling material were placed in a lung phantom filled
with cork chips that simulate lung parenchyma. This phantom was imaged at 1.25
mm collimation, 120 kVp, 100 mA. The CT slices were reconstructed at 1.25 mm, 5
mm, 8 mm, and 10 mm using reconstruction kernels B30, B50, and B70 of a CT
scanner (Volume Zoom, Siemens). The in-plane resolution of these studies was
kept constant at 0.42 mm. For each of these studies, the nodules were segmented
from the helical CT images automatically, and then the volumes were estimated
using three 2D and two 3D methods. All of the 2D methods and one of the 3D
methods are clinically recognized. Another method was developed to estimate the
nodule volumes by taking into account slice overlap information. This 3D method
is based on maximum a posteriori (MAP) estimation. The nodule volumes were
measured by the equivalent volume of displaced water, and varied between 0.03
ml and 1.12 ml. Each nodule volume was automatically estimated using all five
estimation methods for each of the three reconstruction kernels and four slice
thicknesses. The average absolute volume estimation error percentages were
computed.
Results: For 1.25 mm slice thickness and the B50 kernel, the errors were 15.57%
(+-11.84), 32.18% (+-18.87), 23.90% (+-18.22) for the 2D methods, and
22.67%(+-11.65) for the standard 3D method. Using the proposed MAP volume
estimation method, the error was reduced to 7.69% (+-4.83). For 1.25 mm slice
thickness and the MAP estimation method, the errors were 10.41% (+-6.57) and
7.88% (+-5.33) for the B30 and B70 kernels, respectively. For the B50 kernel
and the MAP estimation method, the errors were 32.88% (+-15.95), 61.21%
(+-49.18), and 77.84% (+-63.56) for 5 mm, 8 mm, and 10 mm slice thickness,
respectively.
Conclusion: Volumes of automatically segmented nodules can be accurately
estimated using a thin-slice study with an appropriate kernel and a 3D volume
estimation method. Slice thickness and the volume estimation method have more
influence on the accuracy of the volume estimation than the type of
reconstruction kernel. (M.G., B.A., J.H. are employees of Dag Wormanns. R.H. is
a consultant and S.R. is the President/CEO of CADx Systems, Beavercreek, OH.)
Questions about this event email: mgurcan@cadxsystems.com
Gurcan PhD, M,
Accurate Nodule Volume Estimation from Helical CT Images: Effect of Reconstruction Filter, Slice Thickness, and Volume Estimation Method. Radiological Society of North America 2003 Scientific Assembly and Annual Meeting, November 30 - December 5, 2003 ,Chicago IL.
http://archive.rsna.org/2003/3104662.html