Calibration transfer in mid-Infrared Log Out | Topics | Search
Moderators | Register | Edit Profile

NIR Discussion Forum » Bruce Campbell's List » Calibration transfer » Calibration transfer in mid-Infrared « Previous Next »

Author Message
Top of pagePrevious messageNext messageBottom of page Link to this message

David Cameron (david_cameron)
New member
Username: david_cameron

Post Number: 1
Registered: 3-2011
Posted on Thursday, March 03, 2011 - 8:46 pm:   

Calibration transfer of multivariate models can be extremely difficult in the Mid IR. This is because annomalous dispersion is a major effect in the Mid-Ir. Transmisssion and ATR experiments on homogenous solid and liquid samples are described by the Fresnel equations. These are complex equations. The absorption index is in the real part of the equations, and the refraction index in the imaginary.
In transmission a handwaving description of the spectrum is a constant refraction "spectrum" added to a variable absorption spectrum. Constant because it is an interface effect, variable because it is a pathlength effect. As an example, if you record spectra of a material at pathlengths of 5 and 10 micrometers and peak maxima in the .5 to 1.5 range, you will never be able to cancel them with a subtraction factor of 2. YOu will always have a derivative shaped residual near a band. Ir requires transmission spectra in the 1-25 micrometer pathlength range unless you are into overtones and combination bands. The only way to transfer such calibrations is to use the same pathlength cell, making sure the windows are parallel.
In the case of ATR the spectrum is highly dependent on the angle of incidence. Changing the crystal will change the angle by a small amount and hence change the spectrum. In addition you have the fact that the indicent beam is not collimated, hence you have a range of angles of incidence and hence a range of spectra.
This was mostly covered in papers in the 50s through the 70s.
In summary, it is in the optics, nothing you can do about it except try and calculate it out (which you can, compute the optical constants and use those).
Pragmatically, do what is often so easy in the Mid IR - find a band, and measure it. Call it a PLS prediction if you wish, any PLS result is a dot product between a spectrum and a regression vector, if the regression vector is zero except for one point you mearue a peak height.
And finally, diffuse reflection. You can not get reliable spectra off pure materials, you must grind them down to Ir transmission pathlengths (1-10 um), and dilute them into a ground up matrix. Again, it is all in the Fresenel Equations. However, in this case you have to integrate over all pathlengths and all angles and it is just a mess. You are better off, in my opinion, to press the powder onto a diaond or silicon crystal and do ATR with one of the find accessories which are available.
Top of pagePrevious messageNext messageBottom of page Link to this message

Howard Mark (hlmark)
Senior Member
Username: hlmark

Post Number: 348
Registered: 9-2001
Posted on Thursday, September 02, 2010 - 8:29 am:   

Julio - there are many papers where people try to derive the expected effects from scattering samples, the useful ones are referenced in Don Dahm's book. They all sort of start from the assumption that everyone knows about the "repack" effect, and then go on to try to explain it in terms of some more fundamental considerations. As I said, nobody has solved the problem in general, it always turns out to be too intractable.

More general discussions of NIR measurements and effects encountered are also described in various papers but as is usual in research papers, each one concentrates on the one effect of concern, and you'd have to get at least a dozen (if you can figure out which ones you'd need; most of them are application-oriented so it's not always obvious that there's a fundamental point to be made). Much of the research that has been done deals with variations on the mathematical algorithms that are used in conjunction with NIR.

The bottom line is that if you want a good overview of the optical, mathematical, sample handling and other issues that arise, you'll also have to get a book. The one I can recommend there is the "Handbook of Near-IR Analysis" by Don Burns and Emil Ciurczak, CRC Press (now in its third edition, 2007).

As for NIR instrumentation, the majority of manufacturers of dedicated NIR units sell grating-based instruments. One manufacturer sells a FTNIR unit based on a phase-dispersion interferometer, and several of the manufacturers of mid-IR Michaelson interferometer-based instruments have been adapting them to NIR use.

\o/
/_\
Top of pagePrevious messageNext messageBottom of page Link to this message

Julio Trevisan (lascanter2010)
New member
Username: lascanter2010

Post Number: 4
Registered: 8-2010
Posted on Thursday, September 02, 2010 - 3:47 am:   

Howard - thanks for the book suggestion, I will look for it. Just in case, are there papers on the measurement issues as well? I am glad to hear the description of this 'systematic across wavelengths but random in magnitude and direction' effect. This is something I have experienced a lot just by moving my ATR probe a bit, but could never find anyone mentioning that, to the point I thought this used to happen to no one else.

Jerry - Thanks for your reply. Are NIR instruments usually dispersive, or are FT-NIR instruments common as well? The question of biomarkers is tricky because what I experience is that you can have same classification accuracy using different sub-regions of your spectral range, shedding doubt about where the differences between data classes really are. However, there is always one combination of wavenumbers that yield a slightly better result in terms of classification rate. Thus I currently assume this combination of wavenumbers as being the biomarker. Also, changes are more easily seen in tissue studies than in cytology or cell culture, but this may be because of the difficulties of measuring thin layers of cells on top of slides, thus adding noise to the measurements.
Top of pagePrevious messageNext messageBottom of page Link to this message

Jerry Jin (jcg2000)
Senior Member
Username: jcg2000

Post Number: 33
Registered: 1-2009
Posted on Wednesday, September 01, 2010 - 10:06 am:   

Julio,

You certainly can apply the calibration transfer methods developed for NIR to mid-IR. Instrumentation standardization, as David mentioned, is always an issue, regardless of the type of instrument you use.

FTIR often provides specific absorption band, thus minimizing the need of multivariate calibration as used in NIR and the use of calibration transfer.

FTIR is good at chemical structure elucidation and molecular composition determination. Can you identify the biomarker you need for tissue classification?

Jerry Jin
Top of pagePrevious messageNext messageBottom of page Link to this message

Howard Mark (hlmark)
Senior Member
Username: hlmark

Post Number: 347
Registered: 9-2001
Posted on Tuesday, August 31, 2010 - 6:34 pm:   

Julio - there are several effects of powdered solids, one of them being the non-Beer's law behavior. BTW, if you want to know the latest research into the theory of that, you should read Don Dahm's book: "Interpreting Diffuse Reflectance and Transmittance", NIR Publications (2007). The bottom line to it all is that after more than 100 years of study (counting from Schuster's paper in 1905), there is still no rigorous, ab initio theory of diffuse reflection.

The main experimental artifact caused by the effects of non-homogeneous, scattering, samples is what is variously called "repack effect" or "particle size effect". This the interaction between the light and the inhomogeneities of the sample that cause the measured reflectance to change from one reading to the next in a manner that is systematic across wavelengths, but random in magnitude and direction, from one reading to the next, whenever there is even a small change in the geometry of the measurement (i.e., the sample is moved even a tiny bit).

\o/
/_\
Top of pagePrevious messageNext messageBottom of page Link to this message

Julio Trevisan (lascanter2010)
New member
Username: lascanter2010

Post Number: 3
Registered: 8-2010
Posted on Tuesday, August 31, 2010 - 5:38 pm:   

Hello,

Howard, thanks again for your reply.

"I would also expect that the non-uniform nature of biological tissue would cause problems similar to those found with powdered solids."

What problems do you have in mind? Do you mean non-obedience to Beer's law due to scattering, or is it something else?

I am very keen to understand these instrumental and physical problems.
Top of pagePrevious messageNext messageBottom of page Link to this message

Howard Mark (hlmark)
Senior Member
Username: hlmark

Post Number: 346
Registered: 9-2001
Posted on Tuesday, August 31, 2010 - 2:28 pm:   

Julio - even with the short pathlengths achievable with ATR, unless the samples are dried, water absorbance in the mid-IR is extremely strong and there are "blackout" regions of the spectrum where bands from other components of the sample cannot be seen.

I would also expect that the non-uniform nature of biological tissue would cause problems similar to those found with powdered solids.

My (intended) message was that historically, mid-IR spectroscopists tended to stay away from those types of samples that created the sorts of extraneous problems that the NIR spectroscopists were dealing with; that's what made it have fewer problems, not the non-existence of problem areas. The presence of water in biological samples is a much more difficult problem to deal with in the mid-IR than in the NIR, for example.

Basically, it seems that you're one of the early mid-IR spectroscopists to enter the "deep end of the pool", so to speak, and involve yourself in these issues.

It's also entirely possible, of course, that there are scientists working with mid-IR on tissue and other biological samples, but don't publish in the chemical/spectroscopic literature. You might want to see what you can find by searching the medical literature for work of this nature. Also the patent literature.

\o/
/_\
Top of pagePrevious messageNext messageBottom of page Link to this message

David Russell (russell)
Senior Member
Username: russell

Post Number: 51
Registered: 2-2001
Posted on Tuesday, August 31, 2010 - 1:23 pm:   

Calibration transfer is always an issue.

However, since NIR models often are more complex due to the need to quantitate in the presence of severely overlapping bands it has been a more fertile field for the study of calibration transfer methods.
Top of pagePrevious messageNext messageBottom of page Link to this message

Julio Trevisan (lascanter2010)
New member
Username: lascanter2010

Post Number: 2
Registered: 8-2010
Posted on Tuesday, August 31, 2010 - 1:13 pm:   

Hello,

thanks for your answers.

I am using FT mid-Infrared spectrometry, both transflection and ATR to analyse samples such as cervical cytology, and tissue samples.

A central part of the research is to build data bank to train algorithms to classify unknown data.

From Howard's message I presume that mid-IR presents less problems than NIR and for this reason people don't need to worry about calibration transfer. Is this right?

Actually it seems to me that there are difficulties with mid-IR as well.
Top of pagePrevious messageNext messageBottom of page Link to this message

David Russell (russell)
Senior Member
Username: russell

Post Number: 50
Registered: 2-2001
Posted on Tuesday, August 31, 2010 - 1:11 pm:   

My application was unusual in that the models evolved over many years and much of the variation attributable to the instrument was eliminated through thoughtful calibration augmentation with process data.

Bottom line is that we swapped out an FT instrument with another from the same vendor and the models worked well enough.

I do have gain and bias knobs available but haven't used them. There is only an approximate relationship to regularly run lab reference data and the values have passed that test.
Top of pagePrevious messageNext messageBottom of page Link to this message

Howard Mark (hlmark)
Senior Member
Username: hlmark

Post Number: 345
Registered: 9-2001
Posted on Tuesday, August 31, 2010 - 12:54 pm:   

Dave - did you need to bias/skew correct as is typically done in NIR calibrations? Or use any of the other "tricks" that are typically applied?

If it was a matter of using FT versus grating instruments, then Buchi wouldn't have to do any manipulation of the data, but I'm pretty sure they include those capabilities in their software.

I suspect it's more a matter of the types of samples used. NIR, because of its history, is widely used for measurements of powdered solids or even more unreproducible materials, whereas mid-IR, because of ITS history, is more widely used for clear liquids.

I think that what you said is part of the answer, and that what I said above is another part of the answer, and that there's a third part of the answer, too. The third part is that the mid-IR spectroscopists haven't put themselves in situations, or been put in situations, where they've had to transfer calibrations between different instruments, in cases where the calbration process was "difficult" in the sense that NIR calibrations typically are.

It would help if Julio could tell us something about his samples, though.

\o/
/_\
Top of pagePrevious messageNext messageBottom of page Link to this message

David Russell (russell)
Senior Member
Username: russell

Post Number: 49
Registered: 2-2001
Posted on Tuesday, August 31, 2010 - 12:35 pm:   

Largely because FT based instruments are inherently more reproducible than dispersive instruments.

I've re-used old models built for FT-NIR without modification.
Top of pagePrevious messageNext messageBottom of page Link to this message

Julio Trevisan (lascanter2010)
New member
Username: lascanter2010

Post Number: 1
Registered: 8-2010
Posted on Tuesday, August 31, 2010 - 11:53 am:   

Hello,

I am working with mid-Infrared in a project that aims to collect enough data to train classifier models and this needs calibration transfer because each instrument has its own particularities.

However, all the literature about calibration transfer is applied to Near-Infrared. Could anybody clarify why nobody talks about calibration transfer in Mid-Infrared applications?

Cheers

Add Your Message Here
Posting is currently disabled in this topic. Contact your discussion moderator for more information.