Page 194 - Vibrational Spectroscopic Imaging for Biomedical Applications

170 Cha pte r S i x

6.4.7 Normalization
After a baseline is fitted, many spectra associated with each pixel
have different baseline intensity levels. Normalization is a process
that brings these intensities to the same level thus minimizing
any differences attributed to intensity. When normalizing over
spectra, the norm, or square root of the sum of the squares of the
intensities, is calculated for each pixel and each intensity value in
the given spectrum is divided by that value. This process is known
as vector normalization. Alternatively, peak normalization may be
performed. Using this method, each element or component of a spec-
trum (or pixel in imaging) is divided by a constant that is deter-
mined by the height of a given peak. 56

6.4.8 Smoothing
In conjunction with these standard preprocessing steps, additional
operations may be performed to reduce noise and eliminate other
background features through the process of smoothing. While this
step is often done, it should be carried out with caution, as it may
eliminate minor Raman spectral features that can be utilized in data
analysis steps. Benefits of smoothing include improving the aesthetic
appearance of the spectra, reducing noise (that already may be present
or added during other preprocessing steps), and improving the
efficiency of other operations. Savitsky-Golay filtering is a common
type of technique that uses polynomials to perform the smoothing
operation. The number of data points, polynomial order, and deriv-
ative order are taken into account for the smoothing application. An
alternative to Savitsky-Golay smoothing is gaussian blurring, which is
based on a convolution kernel that is a gaussian function. It is a spa-
tial filter that smoothes sharp contrasts within images. Other convo-
lution filters, such as edge detection or sharpening, may be used to
enhance images. These filters have predefined kernel or filter coeffi-
cients that are used to perform the filtering. Again, precaution must
be taken when applying smoothing and convolution filters to wide-
field Raman images to prevent the removal of spectral features or the
addition of unwanted features.
Once preprocessing is complete, image processing may begin
with various multivariate statistical techniques. Performing these
functions after preprocessing is complete will illustrate the dif-
ferences inherent to the sample without interference from the
background signal.

6.5 Chemometric Analysis of Widefield Raman Images
The images obtained using widefield Raman imaging are spectrally
and spatially resolved images. At every pixel in the image, there is a
spectrum to represent the constituent molecular species contained