Information Technology Reference
In-Depth Information
parameters. Furthermore, in this investigation the surface model obtained after the
statistical instantiation stage was further re
ned by the regularized shape deformation
algorithm. The advantages of integrating this additional stage into the present tech-
nique over other existing attempts to instantiate a patient-speci
c surface model from
a statistical shape model were explained in details in our previous work [
28
]. Brie
y
speaking, such integration enables the present technique to handle more complicated
shape variation of any future instance [
28
].
While accurate, the present approach has limitations related with the number of
training models used to construct the statistical shape models and the number of
validation cases. The accuracy of the present approach depends not only upon how
accurate the image-to-model correspondences can be established but also upon how
well the unknown, patient-speci
fl
c shape variation can be covered by the statistical
shape model that is constructed from a
fixed number of training models. Although
the image-to-model correspondence establishing process has been thoroughly val-
idated in our previous works [
20
22
,
28
] as well as in this investigation, the
statistical shape models used in the present study were constructed from a limited
number of training lumbar vertebral models (39 for the leave-all-in study and 35 for
the leave-four-out study). Furthermore, the validation of the present approach,
though successful, was only conducted on datasets of 3 lumbar vertebrae. Thus, the
results reported in this paper are regarded still preliminary and more thorough
validation study is needed before it can be transferred to a routine usage. None-
theless, the experiment results from the present study demonstrate the ef
-
cacy of
the present approach and the prediction power of the present approach can be
enhanced in the future by incorporating more training models into the statistical
shape model and/or by constructing a patient-oriented statistical shape model.
Acknowledgments The authors gratefully acknowledge the
financial support from the Swiss
National Science Foundation through the National Centers of Competence in Research CO-ME.
The test data are provided by Prof. Dr. S.J. Ferguson.
References
1. Muggleton JM, Allen R (1997) Automatic location of vertebrae in digitized video
fl
fluoroscopic
89
2. Zheng Y, Nixon MS, Allen R (2003) Lumbar spine visualization based on kinematic analysis
from video
images of the lumbar spine. Med Eng Phys 19:77
-
179
3. Wang S, Passias P, Li G, Li G, Wood K (2009) Measurement of vertebral kinematics using
noninvasive image matching method
fl
uoroscopic imaging. Med Eng Phys 25:171
-
E361
4. Bifulco P, Sansone M, Cesarelli M, Allen R, Bracale M (2002) Estimation of out-of-plane
vertebra rotations on radiographic projections using ct data: a simulation study. Med Eng Phys
24:295
-
300
5. Hanson GR, Suggs JS, Freiberg AA, Durbhakula S, Li G (2006) Investigation of in vivo
6DOF total knee arthroplasty kinematics using a dual orthogonal
—
validation and application. Spine 33:E355
-
fluoroscopic system.
J Orthop Res 24:974
-
981
