Information Technology Reference
In-Depth Information
4.3 Fast Reconstruction with Reduced Human Intervention
Producing complete 3D reconstructions based on a large number of manually
identi
ed anatomical landmarks is a long and labor intensive process that is subject
to human error. This task can be greatly accelerated by reducing the amount of
human intervention required for one reconstruction. Section
4.2
provided a hint that
an ef
cient way to do so would be to use a statistical shape model to
nd the most
likely spine given reduced human input.
However, instead of completing partial models, it is possible for a human user to
provide more general cues about the shape of the whole spine. Then, a complete 3D
model can be generated based on these cues. For instance, a human expert can
quickly de
ne a spline that crosses the center of the vertebral bodies from a few
control points on each radiograph. Then, a computer program can
nd the most
likely articulated model corresponding to that input [
29
].
Figure
13
shows an example where a posterior
anterior radiograph and a lateral
radiograph were used. A human expert then selected
-
five or six control points to
de
ne the splines shown in yellow. An articulated model was then computed, and
its associated anatomical landmarks are shown in blue.
The method optimizes a combination of the prior probability of the reconstructed
model and that of the distances between the user-de
ned splines and the centers of
the vertebral bodies. This method is obviously not the only fast reconstruction
method available (see for instance [
22
,
28
,
34
], Chap.
5
, and the references therein).
Fig. 13 Articulated spine models can be used as a priori shape models to perform three-dimensional
spine reconstruction from multiple radiographs. In this example, a spline is defined on each
radiograph, and the most likely articulated model is computed based on an optimization scheme
