Biomedical Engineering Reference
In-Depth Information
The weakness of Raman scattering imposes constraints on the size of the
Raman images that may be conveniently collected. We typically acquire two-
dimensional confocal image planes from an 80
×
60
μ
m spatial region, with
∼
m intervals between acquisition points (357 spectra). It typically requires
2 min to obtain a decent quality spectrum, thus
4
μ
12 h are needed to generate
an image plane under the conditions described above. Although this duration
is too long for certain classes of experiments (e.g., studies of in vivo perme-
ation), it is noted that recent instrumentation advances may permit acquisi-
tion of adequate quality spectra at least 10-100
∼
faster than noted above. It
thus appears that the technical ability to rapidly (seconds or minutes) acquire
large three-dimensional Raman images is imminent.
×
15.2.3 Data Reduction - A Comment About Factor Analysis
As noted above, Raman images of skin are usually derived from hundreds to
thousands of spectra. Examination of each individual spectrum is evidently
impractical. Multivariate approaches are required to condense the information
into a small set of components with a minimum loss of spectral information.
We have found initial evaluation by principal component analysis (PCA) fol-
lowed by factor analysis to be useful for this purpose.
Factor analysis transforms PCA loadings to factor loadings that resem-
ble Raman spectra with scattering intensities of particular components lin-
ear in concentration (all other optical factors being equal, which rarely oc-
curs). Compared to PCA, factor analysis offers a major advantage in that
the loadings resemble real spectra, although they are generally not spectra of
pure components. Thus, since the origin of peaks within the loading (and
therefore in the spectra) is usually available and since spectra structure corre-
lations are occasionally known, molecular structure information from within
the tissue may be acquired. The benefit of the approach to characterize the
molecular structure of skin constituents is demonstrated in some examples
below. However, when comparing factors with spectra it is important to re-
member that factors are generated from all sources of variance in the data.
For example, baseline distortions may appear as “features” in the factor load-
ings which are unrelated to vibrational modes in the sample. To validate the
applicability of factor analysis for a given set of experiments, we routinely
compare factors with appropriately selected spectra, e.g., those from spatial
regions where scores for a particular factor are high. The similarity between
spectra and factor loadings is generally very strong, thereby justifying our
approach.
15.2.4 Delineation of Skin Regions and Accuracy
in their Dimensions from Factor Analysis
As noted above, the thickness of the layered structure of skin is variable across
the tissue, so that no absolute standard is available for estimating uncertain-
ties in Raman-generated measurements of thickness. For the measurements
Search WWH ::
Custom Search