Abstract
A method of reconstructing a three-dimensional
structure from live serial section images taken with
a light or an electron microscope is described. The
method frees researchers from the cumbersome
task of tracing the contour lines of objects. Two
methods of achieving automatic alignment were
tested in order to obtain some technical bases. One
used cross-correlation functions of neighbouring
images; the other, local image features. In the former,
rotational difference and relative displacement
between images were measured at high speed with
the aid of hardware exclusively devoted to the fast
Fourier transform and affine transformations. In
the latter case, techniques used in digital stereophotogrammetry
were employed for identifying corresponding
points in the image pairs. The problem of
image distortion due to local deformations of sections
was solved by using an elastic model where the
image plane was modelled on a material such as a
rubber sheet. With the model a distorted resampling
geometry for the correction was obtained by
numerical calculations under boundary conditions
given by a set of corresponding points in the image
pair. This reconstruction method was applied to a
light micrograph series of minute colonic adenomas
in human familial adenomatosis coli. Some voxel
processings for noise removal and smoothing are
demonstrated. Volumetric presentations and cutting
of the reconstructed data are also demonstrated
in a final three-dimensional rendering.
