Biomedical Engineering Reference
In-Depth Information
second set of 100 registrations (Set 2), the distributions were expanded to [
−
20,
20] degrees and mm.
Since the true registration parameters are unknown, we estimated them
by averaging the manual registration parameters obtained from four clinical
experts. A registration was judged to be successful if the registration parame-
ters were close to the averaged manual values. The thresholds were equal to
the detection thresholds [11], i.e., (4
◦
,
4
◦
,
2
◦
,
2mm
,
2mm
,
3 mm). Note that the
thresholds are tighter than what were used in Zhu [18].
The accuracy and robustness of registrations obtained by minimizing the
similarity measures were analyzed in terms of the mean and standard devia-
tion of the differences between the resultant registration parameters and the
averaged manual registration parameters.
To understand the behavior of these similarity measures in the vicinity
of an optimal registration, the similarity measures were computed in the
neighborhood of manual registration results. It is believed that the manual re-
sults are close to the ground truth. Thus, this calculation would shed light on
the behavior of these measures.
Different strategies to handle the potential numerical instability problem in
this minimization approach are compared by inspecting their similarity function
profiles in the vicinity of the manual registration.
10.5
Results and Discussions
10.5.1
Registration by Cross-Entropy,
Reversed Cross-Entropy, and Symmetric
Divergence Maximization
10.5.1.1
Tx/Em Direct Registration
When the Em images were registered to the Tx images directly, the cross-entropy,
reversed cross-entropy, and symmetric maximization successfully registered 4,
5, and 7 (out of 7) cases, respectively. The results of failed cases by the cross-
entropy and reversed cross-entropy techniques were significantly different from
the intrinsic registrations by visual inspection (the misalignment angles or trans-
lation offsets are too big). The mean and standard deviation of registration
