12 Method Development
293
it is certainly incorrect to assume that the bulk content of a sample will be directly
assignable to the individual spectrum from individual pixels from the image array of
that same sample. In other words, if a sample containing a binary mixture of 92–8
for component A to B, the local aspect of using detector arrays does not result in
every single pixel containing the signal of 92% of A and 8% of B. A distribution
of spectra representing different levels of content for a particular analyte of interest
is obtained for a particular spatial area. In addition, due to the photons’ “random
walk”, the spectrum collected from a single pixel is a weighted averaged signal from
adjacent pixels.
However, this has not stopped scientists from building successful applications
and deploying them in commercial settings. There are two common approaches
to building models for multipoint systems. The first simply consists of taking the
mean of an image as a representative spectrum for the sample being analyzed by
the reference method. The resulting spectrum (average of potentially thousands of
spectra) will have a different signal to noise ratio to the individual pixel spectra but be
relevant for model building. The second approach is to divide the sample in smaller
areas for reference analysis and binning (averaging) spectra spatially so a bin, smaller
than the entire image, can be associated to a local value of the property of interest.
Readers should refer to the chapter on imaging for more information about how these
techniques are used.
12.6 Summary
NIRS is a very powerful technology. It can determine in a matter of seconds the
content of a particular attribute of interest in a complex sample matrix without
preparation, in-line and in real-time. It is sensitive to both the physical and chemical
make-up of the sample which can make any deployment challenging. The Analytical
Quality by Design framework can significantly streamline method development and
lifecycle management efforts to ensure a successful deployment and long-term value
generation from a NIR spectroscopic method. The process of stating the method goals
and requirements, performing feasibility study(ies), and a risk assessment before
method development will help address knowledge gaps and set expectations before
significant resources are spent. The result is a robust method that can be tested and
validated if required by the regulatory environment of use. Continuous improvement
through method lifecycle management ensures method performance over the useful
life of the method.
293
it is certainly incorrect to assume that the bulk content of a sample will be directly
assignable to the individual spectrum from individual pixels from the image array of
that same sample. In other words, if a sample containing a binary mixture of 92–8
for component A to B, the local aspect of using detector arrays does not result in
every single pixel containing the signal of 92% of A and 8% of B. A distribution
of spectra representing different levels of content for a particular analyte of interest
is obtained for a particular spatial area. In addition, due to the photons’ “random
walk”, the spectrum collected from a single pixel is a weighted averaged signal from
adjacent pixels.
However, this has not stopped scientists from building successful applications
and deploying them in commercial settings. There are two common approaches
to building models for multipoint systems. The first simply consists of taking the
mean of an image as a representative spectrum for the sample being analyzed by
the reference method. The resulting spectrum (average of potentially thousands of
spectra) will have a different signal to noise ratio to the individual pixel spectra but be
relevant for model building. The second approach is to divide the sample in smaller
areas for reference analysis and binning (averaging) spectra spatially so a bin, smaller
than the entire image, can be associated to a local value of the property of interest.
Readers should refer to the chapter on imaging for more information about how these
techniques are used.
12.6 Summary
NIRS is a very powerful technology. It can determine in a matter of seconds the
content of a particular attribute of interest in a complex sample matrix without
preparation, in-line and in real-time. It is sensitive to both the physical and chemical
make-up of the sample which can make any deployment challenging. The Analytical
Quality by Design framework can significantly streamline method development and
lifecycle management efforts to ensure a successful deployment and long-term value
generation from a NIR spectroscopic method. The process of stating the method goals
and requirements, performing feasibility study(ies), and a risk assessment before
method development will help address knowledge gaps and set expectations before
significant resources are spent. The result is a robust method that can be tested and
validated if required by the regulatory environment of use. Continuous improvement
through method lifecycle management ensures method performance over the useful
life of the method.