Introduction - High-Throughput Image Reconstruction and Analysis - page 3

necessarily tied to microscopy, but whose ideas and techniques are quite general
and very relevant.
1.1 Part I: Emerging Technologies to Understand Biological Systems
In Part I, we examine new technologies to investigate biological systems. These
enable unprecendented levels of observation in spatial and temporal detail.
1.1.1 Knife-Edge Scanning Microscopy: High-Throughput Imaging and
Analysis of Massive Volumes of Biological Microstructures
In Chapter 2, Choe, Abbott, Han, Huang, Keyser, Kwon, Mayerich, Melek, and
McCormick describe a unique apparatus that combines sectioning and imaging
of biological tissue. This apparatus is suitable for acquiring images of entire or-
gans, such as a plastic-embedded mouse brain, at a resolution of approximately
0.6 m/pixel. This results in a total data size of approximately 20 terabytes. The
authors have developed vector tracking methods to process these images to extract
meaningful structures such as vascular or neural filaments. They exploit parallel
algorithms in order to address the high computation required. Finally, they employ
special techniques from the field of visualization, such as streamlines, in order to
be able to render the large data sets effectively.
1.1.2 4D Imaging of Multicomponent Biological Systems
In Chapter 3, Palaniappan, Bunyak, Nath, and Goffeney investigate time-varying
imaging of biological entities such as cells in motion. This provides important infor-
mation about temporal characteristics, which can be used to understand phenom-
ena such as cell growth and development. This chapter describes the application
of graphical processing units (GPUs) to overcome the computational bottleneck.
1.1.3 Utilizing Parallel Processing in Computational Biology Applications
In Chapter 4, Wagner and Jordan discuss an important problem in the field of
computational biology: the modeling of tumor growth. A mathematical model
is formulated that captures essential processes governing tumor state, including
creation, migration, and death, and the interactions of tumor cells with a tissue
environment, including oxygen consumption and enzyme production. The behavior
of the model is probed through a discrete-time simulation. The challenge is to be
able to apply the model to a biologically realistic number of tumor cells, in order
to be clinically relevant. This turns out to be of the order of a billion tumor cells.
The authors show that this is computationally tractable through parallelization
techniques. Their techniques, as discussed in the section on domain decomposition,
are quite general and can be applied to other problems, such as the tracking of
moving cells in 4D microscopy applications. Other applications, such as contour
tracking in 3D image stacks, can also benefit from the domain decomposition
techniques described in this chapter, as it concerns changes in the process that
owns a cell's computations as the cell migrates. A similar situation occurs when
the contour of a cell shifts as one traces it along successive sections.