Knife-Edge Scanning Microscopy: High-Throughput Imaging and Analysis of Massive Volumes of Biological Microstructures - High-Throughput Image Reconstruction and Analysis - page 16

(multiple times) during processing. One way to overcome this problem is to visit
only a subset of relevant voxels.
One of the latest approaches addressing the above issues is the vector tracking
algorithm [20, 21], where only local information around the object of interest is
examined, thus avoiding visiting every voxel in the volume data set. The basic idea
is to begin with a seed point, estimate the trajectory of a fiber by means of template
matching. Subsequently, small steps are taken along the fiber, continually adjusting
the current position based on the estimated trajectory. Our group at Texas A&M
generalized the approach for more robust tracking [22], which we will describe in
detail in Sections 2.3 and 2.4 (also see [23], which uses level sets).

2.2 Knife-Edge Scanning Microscopy

The instrument comprises four major subsystems: (1) precision positioning stage
(Aerotech), (2) microscope/knife assembly (Micro Star Technology), (3) image cap-
ture system (Dalsa), and (4) cluster computer (Dell). The specimen, a whole mouse
brain, is embedded in a plastic block and mounted atop a three-axis precision stage.
A custom diamond knife, rigidly mounted to a massive granite bridge overhang-
ing the three-axis stage, cuts consecutive thin serial sections from the block. Unlike
block face scanning, the KESM concurrently cuts and images (under water) the tis-
sue ribbon as it advances over the leading edge of the diamond knife. A white light
source illuminates the rear of the diamond knife, providing illumination at the lead-
ing edge of the diamond knife with a strip of intense illumination reflected from the
beveled knife-edge, as illustrated in Figure 2.2. Thus, the diamond knife performs
two distinct functions: as an optical prism in the collimation system, and as the
tool for physically cutting thin serial sections. The microscope objective, aligned
perpendicular to the top facet of the knife, images the transmitted light. A high-
sensitivity line-scan camera repeatedly samples the newly cut thin section at the
knife-edge, prior to subsequent major deformation of the tissue ribbon after imag-
ing. The imaged stripe is a 20-mm-wide band located at the bevel at the very tip of
the diamond knife, spanning the entire width of the knife. Finally, the digital video
signal is passed through image acquisition boards and stored for subsequent anal-
ysis in a small dedicated computing server. The current server is a dual processor
PC (3.2 GHz/2MB Cache, Xeon) with 6 GB of memory, built-in 1-TB storage, con-
nected to an archival RAID attachment. The process of sectioning and imaging is
fully automated with minimal human intervention. Figure 2.3 shows a screen-shot
of the stage controller/imaging application developed and maintained in our lab.
A quick calculation puts us in context, regarding the massiveness of the data
that KESM can produce. Consider the acquisition of volume data representing a
plastic-embedded mouse brain (15 mm Anterior-Posterior, 12 mm Medial-Lateral,
6 mm Dorsal-Ventral). A 40X objective has a field of view (knife width) of
0.625 mm. Sixteen strips (each 0.625 mm wide by 15 mm long) are cut for
each z -axis section (like plowing a field). For a ( z -axis) block height of 6 mm,
12,000 sections must be cut, each 0.5 mm thick. The integrated tissue ribbon length
(15 mm/strip

× 16 strips/section × 12,000 sections/mouse brain) is 2.9 km. The