NeoDym has studied the
complete end-to-end modeling of an instrument which
will determine the level of glucose in blood by
measuring the infra-red absorption spectrum.
Infra-red molecular absorption
spectra do not have sharp spectral lines like the
spectra from many atoms. They are very broad and the
peaks all run into each other. Water is also very
absorptive in the near infra-red and in the region
where glucose is active. This is at a wavelength of
just under 1000 nanometers wavelength.
The simulation started by
modeling the spectral output of the quartz halogen
lamp, modeling the absorption of the blood with 10
main analytes present and then modeling the
detector/electronics and the data reduction process.
From the simulation we were able to determine the
major contributors to error in the glucose estimate
and how they interacted.
Two types of data reduction
were modelled - i.e. Partial Least Squares (PLS) and
Classical Least Squares. PLS requires no assumptions
about the spectra of the various analytes, but does
require extensive calibration and 'training' of the
PLS algorithm. CLS, on the other hand, utilises the
known spectra of the blood analytes, but does not
require a calibration or training process.