Page 318 - Vibrational Spectroscopic Imaging for Biomedical Applications

292 Cha pte r Ni ne

Raman spectroscopic imaging perfectly combines spatial and spec-
troscopic (biochemical) information and enables locating the origin of
macroscopic Raman tissue signals and changes therein caused by dis-
ease. Biomedical application of Raman imaging is currently for a
large part focused on more fundamental cancer research, where
Raman spectroscopic imaging has shown to be a powerful research
tool that can provide molecular information at subcellular resolution
that is difficult to obtain by other methods. In this role its results can
be used to develop optimal probes 10,43,73,78,79 and data-analysis models
that can be applied clinically in vivo. The fact that acquisition of these
maps may take hours to days is of minor importance.
For clinical in vitro application, Raman imaging should overcome
its main limiting factor, the time needed to generate images that is use-
ful in clinical diagnosis. For tissue section screening applications, one
would like to have diffraction-limited spatial resolution (0.5 μm, or
even higher using the accuracy of a motorized sample stage) in a total
2
field of view in the order of a cm . This would result in a total image of
20.000 × 20.000 pixels, i.e., 400 million pixels. With a point mapping
technique and 1 second signal collection time, this would take about
12.7 years. Most of this image is uninformative and will never be
inspected, at least not in that resolution. When a pathologist inspects a
stained tissue section, he starts with a low magnification, i.e., a low-
spatial-resolution image (in the order of 50 μm). He/she then selects
areas for further inspection at a higher magnification (order of 5 μm).
Only if he wants to examine the way a cell is build up, he would use the
highest resolution, i.e., 0.5 μm. This zooming ability is typically what an
useful Raman imaging instrument also would need to provide. This
means the ability to generate images in the order of 200 × 200 pixels, i.e.,
40 thousand pixels with a pixel size of 0.5, 5, or 50 μm. With a point
mapping technique and 1 second signal collection time this would take
about 11.1 hours per image which is way long for practical applications.
By using a line imaging technique to measure 128 pixels at a time, and
by bringing the signal acquisition time back to 0.2 second (and make
full use of noise filtering aspects of multivariate techniques), this time
can be reduced to somewhat more than a minute, which comes close to
something practical.
There are several ways to bridge this gap from research tool to
clinical application. The instrumentation can still be improved and
extended. 80–82 The amount of light that can be focused on a point is
limited because of photodamage to the tissue. However, if spread,
higher-laser powers can be used. New EMCCDs can help to lower the
needed CCD readout time which is especially useful when short-
acquisition times are used. Multiplexing techniques like Hadamard
transform imaging can also help to reduce the influence of detector
noise, but at the cost of not detecting all generated Raman photons.
With respect to data analysis of biomedical Raman images there
is still much room for progress. Robust data-analysis techniques that