Page 156 - Vibrational Spectroscopic Imaging for Biomedical Applications

132 Cha pte r F i v e

• Separation between the strong absorbances in the spectrum;
this is preferable but not essential, as several numerical tools,
carefully applied to sufficiently large datasets, can be
employed to overcome some of these issues.
• Physical size and distribution of localized fractions.

5.3.1 Spectral Resolution
Biologically important functional groups produce characteristic,
well-separated peaks. 35 Grey and white matter regions of brain
tissue are readily distinguished in vibrational spectra, because of
the significant difference in the amount of membrane (phospholipid
bilayer) present. White matter axons are shielded by a sheath of myelin
that is a manyfold thickness of cell membrane. The intense bands at
2800 to 2950 cm are characteristic of lipid acyl chains: the dominant
−1
−1
peaks at 2848 and 2926 cm arise from the symmetric and asymmetric
stretch of CH , while the weaker peaks at 2875 and 2950 cm are from
−1
2
CH stretching of the terminal methyl groups. 42–44 A small absorption
3
due to the stretch of a CH bond on an unsaturated carbon (=C⎯H)
−1
may be observed at 3012 cm . Amide groups within an α-helical pro-
tein will absorb radiation at 1655 to 1660 cm , whereas, when β-sheet
−1
amide groups are present, the maximum absorption occurs at lower
−1 45
energy, about ~1630 cm .
Our IR data are typically acquired at a nominal spectral resolu-
−1
tion of 4 cm ; absorbance features that are spectrally separated by
less than this amount will not be resolved. The absorbance bands of
biological molecules are much broader than this limit. The apparently
simple bands observed in the IR spectra of tissue (e.g., Fig. 2 in Ref. 45)
are really the summation of many, often broad, overlapping absorbance
bands centered at nearly identical wavenumbers. Numerical tech-
niques for artificially enhancing the spectral resolution are sometimes
employed in these cases; however, to date we have based our analy-
ses on the data as originally recorded. Spectral profiles and sophisti-
cated data analysis algorithms permit classification of large datasets
according to small variations in spectral profile. 46–48 These analyses,
both supervised and unsupervised clustering algorithms, as well
as artificial neural networks, are leading to new applications for
rapid recognition of everything from bacteria to cancer.
Spectral resolution in Raman analyses matches the limit in IR, at
–1 29,30,37
∼1 to 2 cm . Raman spectra of coronary artery have been used
to quantify chemical composition of coronary plaques, in terms of
cholesterol, cholesterol esters, triglycerides and phospholipids, and
calcium salts. 38,49 Multivariate analysis has been used to improve
spectral analysis and decomposition of data according to component
fractions. We have used Raman microspectroscopy to verify the
identity of crystal depositions as calcium hydroxyapatite in cardio-
myopathic hamster heart tissue sections. 50