446
H. M. Heise
There have been studies by other groups by comparing the performance of NIR
versus Raman spectroscopy assays by using solutions of glucose, lactate and urea in
aqueous phosphate buffer. Surprisingly, the NIR-assay outperformed that by using
Raman spectroscopy. Standard errors of prediction were 0.24, 0.11 and 0.14 mmol l
−1
for glucose, lactate and urea, respectively, from near-IR spectra, while SEP values
of 0.40, 0.42 and 0.36 mmol l
−1 were achieved with Raman calibration modeling
for glucose, lactate and urea, respectively. Differences between instrumental signalto-noise ratios were responsible for the better performance of the near-IR spectrometer. It is certainly an advantage that NIR spectra can be measured by transmission cells with quartz or glass windows or inside glass vials or less often by total
reflectance using NIR spectrometers that is more affordable than MIR or Raman
instrumentation [13].
Assay selectivity has been and is still in the focus of many analytical spectroscopists, and one valuable approach has been developed by using the net analyte
signal for investigating the non-overlapping spectral features of individual analytes.
Arnold et al. [14] investigated the selectivity of near-infrared spectroscopy for the
independent measurement of glucose, glucose-6-phosphate and pyruvate in ternary
mixtures under physiological pH conditions. Spectral data from multiple measurements within the combination band region had been collected and multivariate calibration modeling was carried out using the PLS method. Despite the high similarity of the individual spectra, the principal conclusion was that selective analytical
measurements were possible for the three analytes, based on high quality long-wave
near-infrared transmission spectra.
Another important analyte in clinical chemistry is hemoglobin, whose concentration is commonly used in clinical medicine to diagnose anemia, identify bleeding,
and for managing red blood cell transfusions. Automated hematology analyzers are
usually applied for quantitative analysis. Usual reference methods for hemoglobin
are not reagent-free in contrast to an NIR spectroscopic assay. Besides the use of
traditional laboratory equipment, anemia diagnosis can also be accomplished by
quantifying the hemoglobin (Hb) concentration with point-of-care testing (POCT)
devices such as the HemoCue test systems (Ängelholm, Sweden) [15]. In the past,
several papers have been published on hemoglobin quantification with NIR spectrometry; see for example [16]. Here, the combination band region was considered
also for wavelength variable selection using multiple linear regression (MLR) with
linear summation equations based on three and four characteristic wavelengths. The
cross-validated standard error of prediction (SEP) for hemoglobin was 1.25 g dl
−1
with a four term model over the concentration range from 5.9 to 20 g dl
−1 . A recently
presented development is based on using a laptop-controlled NIR spectrometer (via
USB interface, spectral range 900–1700 nm [11,100–5900 cm
−1 ]). The device had
been connected to a supercontinuum broadband white light laser source by using
an optical fiber bundle. Transmission cuvettes were of 5 mm optical pathlength.
Prediction errors from a validation sample population were around 0.44 g dl
−1 ,
either with full spectrum use or with different variable selection methods, using
three or two wavelengths only. An informative literature overview on past in vitro
and non-invasive approaches is also given [17].
H. M. Heise
There have been studies by other groups by comparing the performance of NIR
versus Raman spectroscopy assays by using solutions of glucose, lactate and urea in
aqueous phosphate buffer. Surprisingly, the NIR-assay outperformed that by using
Raman spectroscopy. Standard errors of prediction were 0.24, 0.11 and 0.14 mmol l
−1
for glucose, lactate and urea, respectively, from near-IR spectra, while SEP values
of 0.40, 0.42 and 0.36 mmol l
−1 were achieved with Raman calibration modeling
for glucose, lactate and urea, respectively. Differences between instrumental signalto-noise ratios were responsible for the better performance of the near-IR spectrometer. It is certainly an advantage that NIR spectra can be measured by transmission cells with quartz or glass windows or inside glass vials or less often by total
reflectance using NIR spectrometers that is more affordable than MIR or Raman
instrumentation [13].
Assay selectivity has been and is still in the focus of many analytical spectroscopists, and one valuable approach has been developed by using the net analyte
signal for investigating the non-overlapping spectral features of individual analytes.
Arnold et al. [14] investigated the selectivity of near-infrared spectroscopy for the
independent measurement of glucose, glucose-6-phosphate and pyruvate in ternary
mixtures under physiological pH conditions. Spectral data from multiple measurements within the combination band region had been collected and multivariate calibration modeling was carried out using the PLS method. Despite the high similarity of the individual spectra, the principal conclusion was that selective analytical
measurements were possible for the three analytes, based on high quality long-wave
near-infrared transmission spectra.
Another important analyte in clinical chemistry is hemoglobin, whose concentration is commonly used in clinical medicine to diagnose anemia, identify bleeding,
and for managing red blood cell transfusions. Automated hematology analyzers are
usually applied for quantitative analysis. Usual reference methods for hemoglobin
are not reagent-free in contrast to an NIR spectroscopic assay. Besides the use of
traditional laboratory equipment, anemia diagnosis can also be accomplished by
quantifying the hemoglobin (Hb) concentration with point-of-care testing (POCT)
devices such as the HemoCue test systems (Ängelholm, Sweden) [15]. In the past,
several papers have been published on hemoglobin quantification with NIR spectrometry; see for example [16]. Here, the combination band region was considered
also for wavelength variable selection using multiple linear regression (MLR) with
linear summation equations based on three and four characteristic wavelengths. The
cross-validated standard error of prediction (SEP) for hemoglobin was 1.25 g dl
−1
with a four term model over the concentration range from 5.9 to 20 g dl
−1 . A recently
presented development is based on using a laptop-controlled NIR spectrometer (via
USB interface, spectral range 900–1700 nm [11,100–5900 cm
−1 ]). The device had
been connected to a supercontinuum broadband white light laser source by using
an optical fiber bundle. Transmission cuvettes were of 5 mm optical pathlength.
Prediction errors from a validation sample population were around 0.44 g dl
−1 ,
either with full spectrum use or with different variable selection methods, using
three or two wavelengths only. An informative literature overview on past in vitro
and non-invasive approaches is also given [17].