Initial validation studies have been conducted to compare the single- and multi-level registration strategies [9] and to assess the usefulness of a volume-preservation constraint [10].
METHOD
Three images pairs, where no motion was visible between pre- and post-contrast image, were selected from a large database. For each patient, four deformations were generated and target registration errors were determined as described in [7].
RESULTS
(a)
(b)
(c)
(d)
Figure 1: Target registration error (TRE)
after simulation, rigid registration
and multi-resolution non-rigid registration
with control point spacing of 20mm, 20mm->10mm and
20mm->10mm->5mm.
TRE was evaluated over (a,b) the
whole breast tissue and (c,d) the region of the enhancing
lesion for (a,c) the unconstrained non-rigid registration and (b,d)
the volume-preserving non-rigid registration.
The graph shows the mean and standard deviation of TRE for three
patients and four deformations. (from [10])
CONCLUSIONS
We found that the volume preserving non-rigid registration was more accurate than the unconstrained method. Severe local deformations were better compensated by finer control point spacing. However, the contrast enhancing lesions were more accurately aligned at a control point spacing of 20mm.
References: