![[*]](/icons/latex2html/foot_motif.gif){kind=icon}
.
Two full-3D neck volumes were provided, neckD
and neckI. NeckI was of the chin down and neckD was with the head flexed backwards within the confines of the scanner bore.
Both were of the same subject.
Since the imaging field of view had been the spinal column specifically, the volumes did not extend to include the whole neck laterally.
The images exhibited a strong intensity ramp, with high values at the back of the neck and total signal loss in the face. We pre-processed the images to give a more uniform range of intensities in the anterior-posterior direction, using the scheme described by Fig. 7. Due to memory capacity and time constraints, it was not feasible to apply the fluid registrations directly to the full-resolution data sets; hence they were downsampled, by blurring with a Gaussian of standard deviation
and storing alternate pixels (Fig. 7, right). Since the Gaussian blurring for downsampling, and intensity gradient calculations during registration, were performed in the Fourier domain and required image dimensions of powers of two, we used zero padding to give full resolution dimensions
and downsampled volumes of
.
![\begin{figure}\makebox[\textwidth]{
\epsfxsize=0.2\textwidth \epsfbox{Dorig19of3...
...x{Dmask19of38.ps}
\epsfxsize=0.2\textwidth \epsfbox{Dhf16of32.ps}
}
\end{figure}](img49.gif){kind=small} |
We segmented the spinal vertebrae slice-by-slice from the full-resolution source volumes using the xdispunc display tools. The contrast between vertebrae and intervening tissue was variable and so segmentation was performed manually with reference to an atlas [13].
The segmentations were converted to
binary spine volumes which were then downsampled using the same process as for the necks.
Both fluid registration tests (UF, MF3)
were applied in a
six-level scale space (Gaussian blurs of spatial standard deviation 2i with
). Within each scale level, the fluid was set to iterate through at least three timesteps, with an optional extra 100 timesteps until the stopping criterion was met, the stopping criterion being a reduction in correlation coefficient of less than 10-4.
On termination, we upsampled the displacement fields obtained from both fluid registrations of neckD to neckI and applied them to the original neckD images, to give full-resolution deformations. These are shown in Fig. 8 (top, centre).
![\begin{figure}
\makebox[\textwidth]{
\epsfysize=0.2\textwidth \epsfbox{D32.ps}
\...
...sfbox{vDtoIfrI16.ps}
\epsfxsize=0.2\textwidth \epsfbox{blank.eps}
}
\end{figure}](img51.gif){kind=small} |
We applied the transformation fields produced by
both fluid tests to the intial spine volumes segmented from neckD;
the results of the volume-rendered spines are shown from two angles in
Fig. 9.
The UF registration shows an extension of the upper two vertebrae of the spines on comparison to the original segmentation from neckD (Fig. 9 far right).
![\begin{figure}\makebox[\textwidth]{
\epsfxsize=0.10\textwidth \epsfbox{vgs_Dspin...
....15\textwidth]{
\epsfysize=0.182\textwidth \epsfbox{var_bit.ps}
}
}
\end{figure}](img52.gif){kind=small} |
Figure 10
shows logs of the Laplacian and elasticity energies as local deformation metrics computed from the displacements of both registrations of neckD
to neckI.
The deformation metrics clearly show dark patches in the vertebrae in MF3 indicating low distortion: compare Fig. 10 (left) with those of the UF (Fig. 10 right).
![\begin{figure}\makebox[\textwidth]{
\epsfxsize=0.18\textwidth \epsfbox{fDtoIlLa1...
...{vDtoIlLa16.ps}
\epsfxsize=0.18\textwidth \epsfbox{vDtoIlel16.ps}
}
\end{figure}](img53.gif){kind=small} |
volume; after division pixelwise by the same image blurred with a Gaussian of spatial standard deviation
;
masked with the aid of the automatic contouring and manual alteration tool in xdispunc; downsampled by a half