Scalable Image Registration and 3D Reconstruction at Microscopic Resolution - High-Throughput Image Reconstruction and Analysis - page 200

Figure 8.9 (a) A sample 3D reconstruction of mouse placenta with virtually excised corner. Only
a fraction of the reconstructed volume is shown at high resolution due to the limitations of the
rendering software. (b--e) Registration of mouse mammary images: (b) a 1,000 × 1,000-pixel patch
from the base image; (c) corresponding 1,000 × 1,000-pixel patch taken from the float image;
(d) patch from (c) after nonrigid transformation of the float image; (e) overlay between (b) and (d),
with the grayscale representations embedded in the red and green channels respectively.
image area for both the mammary and placenta datasets at the respective smallest
values of W 1 . Due to differences in content sparsity between the two datasets, the
placenta images have nearly twice the intensity feature density of the mammary
dataset.
Figure 8.10 (a) Overlay of base and float placenta images after rigid registration. (b) High-
resolution differenced patch from (a). (c) Overlay of the base and float images after nonrigid regis-
tration. (d) High-resolution differenced patch from the panel (c). (e, f ) Rendering of an edge view
of placenta reconstruction, the frontal views represent virtual cross-sections of the reconstructed
tissue: (e) with rigid registration alone, no coherent structures are apparent in the frontal view;
(f ) nonrigid registration corrects the distortions apparent in (e) and the reconstructed volume is
then suitable for further analyses.