Introduction - High-Throughput Image Reconstruction and Analysis - page 6

processed by the same location. This has been an open question in the literature,
with studies supporting both hypotheses. Xiao and Kaplan designed a technique to
measure the cortical response to stimuli that differed only in orientation or color.
By using support vector machines to train a classifier, they were able to show that
certain areas of the cortex could predict the input stimulus with a high degree of
accuracy, indicating they were coding the stimulus. Though the problem can be
thus formulated in a mathematically elegant fashion, it is very computationally
demanding, as separate classifiers need to be designed for every local region of the
cortex. They overcame this by parallelizing the computation, bringing down the
time from a few years to a few hours. Overcoming the computational bottleneck
allows scientists the ability to think about and formulate their problem in the most
appropriate manner.
1.3.4 High-Throughput Analysis of Microdissected Tissue Samples
In Chapter 11, Rodriguez-Canales, Hanson, Tangrea, Mukherjee, Erickson, Albert,
Majumdar, Bonner, Pohida, Emmert-Buck, and Chuaqui describe exciting new
developments in applying molecular techniques to tissue samples for studying dis-
eases. A technique that is being increasingly used is tissue microdissection, which
requires advances in multiple fields, including histopathology, image analysis,
and microscopy instrumentation. This technique opens new directions of inves-
tigation in the biological sciences. This chapter provides important background
to understand innovative techniques that are being applied to probe and isolate
relevant microscopic cellular features from biological samples. The advantages
and limitations of these techniques are presented, which should be valuable to
practitioners interested in applying image processing techniques to this domain.
An emerging application is the use of fluorescent labeling to identify different cell
populations. This poses several challenges in image analysis, such as segmentation.
1.3.5 Applications of High-Performance Computing to Functional Magnetic
Resonance Imaging (fMRI) Data
In Chapter 12, Garg examines the analysis of fMRI images, an area of consid-
erable interest in neuroscience, as it provides data on a whole, behaving brain
in human subjects. Due to the large number of spatial voxels and long temporal
durations used in typical experiments, the size of the data gathered can be enor-
mous, especially if multiple subjects are considered. As a result, researchers have
focused on utilizing less computationally demanding techniques such as the gen-
eral linear model. However, more sophisticated modeling of the temporal aspects
of the signals gathered leads to a richer interpretation of the data, including the
establishment of networks of related voxels, and directed graphs that represent
causal relationships between voxels. This constitutes a paradigm shift in the anal-
ysis of fMRI images, and enables a wider variety of experimental protocols to be
adopted. The price to be paid for this increased capability is the increased compu-
tation involved. Garg shows how the effective parallelization of these techniques
renders tractable the more sophisticated mathematical modeling techniques such
as Granger causality analysis.