Biomedical Engineering Reference
In-Depth Information
Fig. 16.6.
Representative Raman spectra, after background subtraction, of blood
serum (
above
) and urine (
below
). Spectra from multiple subjects are shown (data
replotted from [5]). The spectrum of blood serum contains many chemical signatures
at comparable amplitudes, whereas that of urine is dominated by a urea peak
acquired both before and after a filtration step that removed large molecules
(such as proteins). They showed that this strongly reduced the background
fluorescence and improved the accuracy with which lighter molecular species,
such as glucose, were predicted.
In the decade since these first reports, various advances in blood serum
Raman spectroscopy have been reported. As mentioned above, Qu developed
a system with a waveguiding sample-filled capillary tube and two linear arrays
of abutting fibers, with 100 mW sample illumination at 745 nm. On synthetic
mixtures that simulated human serum, they obtained prediction errors for
glucose and urea of 0.5 and 0.4 mM, respectively [4].
The most extensive test to date was performed by Rohleder et al. [19, 20]
using 400 mW of 785 nm excitation (200 mW at sample), a quartz cuvette,
epi-detection, 5 min of spectral integration, and PLS calibration. The number
of blood donors was 247, much higher than previous studies. Concentrations
of glucose, urea, uric acid, cholesterol, triglycerides, and both high-density and
low-density lipoprotein (HDL and LDL) were predicted with significant cor-
relations. For the smaller molecules, ultrafiltration significantly improved the
prediction accuracy. Concentration errors for glucose and urea were 0.95 and
0.73 mM, respectively, when predicted in native serum; these values improved
to 0.38 and 0.35 mM with ultrafiltration.
Recently, Qi et al. made measurements on approximately 70 serum samples
using an epi-illuminated waveguiding sample chamber, 160 mW of 830 nm ex-
citation,
up
to
150 s
of
spectral
integration,
and
PLS
leave-one-out
Search WWH ::
Custom Search