Biomedical Engineering Reference
In-Depth Information
8.4 Data Representation and Result Visualization
Visualization of results, their statistical validity, and analysis of results are
becoming increasingly important. Several techniques are available to represent
data and compact representations are increasingly being made by images. The
ease of visualization, however, must always be tempered with the need to
present data completely and in an accurate manner. In particular, two areas
are of relevant important. The first is two-dimensional spectral representation
to better understand spectral structure while the second is more useful for
imaging.
8.4.1 Two-Dimensional Correlation Analysis
Two-dimensional data analysis [160, 161] is a very powerful technique that ex-
amines correlated changes in spectra with changes in any other measurement
of sample perturbation. The elegance of the method demonstrated in earlier
studies has been extensively augmented by finer details on application and
numerous examples. The primary advantage of two-dimensional correlation
analysis lies in the extension of data examination “space” to a second do-
main. Subtle changes that may not be easily discernable in spectra and even
weak spectral effects may be easily enhanced and understood in the context
of molecular spectra [162].
8.4.2 Morphologic-Spectroscopic Data Representation
While correlations between clusters and groups of spectra can be easily vi-
sualized in the form of images, spectroscopic imaging data are becoming in-
creasingly prevalent. Spectral-spatial associations are both easy to visualize
and can be compactly represented in imaging format using a number of tech-
niques for evaluating, representing, and enhancing images from spectroscopic
imaging data [163].
8.5 Case Study: Classification of Prostate Tissue
In this section, we present the development of an automated protocol for
prostate tissue histology [164] from infrared spectroscopic imaging data as
an example of the techniques described (Fig. 8.11). The data is three dimen-
sional with
x
-
y
axes representing the image plane and the
z
-axis representing
the spectral dimension. After data acquisition, two important pre-processing
steps, namely baseline correction and de-noising, are performed. Since the
entire data set is derived from human tissue samples, the spectra have sim-
ilar characteristics and, therefore, a manually chosen set of pre-defined wave
number could be used as the reference points for baseline correction. It is
−
Search WWH ::
Custom Search