Microscopic Slice Image Processing and Automatic Labeling - Biomedical Signal Analysis: Contemporary Methods and Applications - page 349

13

Microscopic Slice Image Processing and Automatic
Labeling

A supervised interpretation of the initial data analysis model from
section 4.1 leads to a classification problem: given a set of input-output
samples, find a map that interpolates these samples, and, hopefully
generalizes well to new input samples. Such a map thus serves as classifier
if the output consists of discrete labels. Classification based on support
vector machines [36, 37, 229] or neural networks [111] has prominent
applications in biomedical data analysis. Here we review an application
to biomedical image processing [260].
While many different tissues of the mammalian organism are capable
of renewing themselves after damage, it was long believed that the
nervous system is not able to regenerate at all. Nevertheless, the first
data showing, that the generation of new nerve cells in the adult brain
could happen were presented in the 1960s [7], showing new neurons in
the brain of adult rats. In order to quantify neurogenesis in animals,
newborn cells are labeled with specific markers such as BrdU; in brain
sections these cells can later be analyzed and counted through the use of
a confocal microscope. However, so far this counting process had been
performed manually.
The goal of this chapter is to automate the task of counting labeled
cells, which is currently done manually in many laboratories. Our novel
algorithm contributes to a substantial speed-up in experimental settings.
Furthermore, when comparing manual counts, differences in the counts
are often noticed; hence, with an automated counting algorithm we hope
to achieve an objective counter with known error bounds.
The chapter is organized as follows: section 13.1 presents the nec-
essary neurobiological background of the analyzed section images. We
then give an overview of the ZANE cell-counting algorithm in section
13.2. Section 13.3 presents an ecient algorithm for image stitching used
in ZANE to allow for counting larger brain sections. The neural-network
cell classifier is constructed in section 13.4, and is then used to analyze
cell images in section 13.5. Comparisons with other methods are pre-
sented in section 13.6, and our main results are shown in section 13.7,
comparing ZANE with manually counted section images. We finish with
a discussion of further applications and future work in section 13.8.