Tip:
Highlight text to annotate it
X
[ Music ]
>> Troy Ernst: Well, I think I've got the easiest sounding
title here, but then I fill it with a lot of tongue twisters
to make it interesting.
So my presentation deals with signal to noise ratios
and micro XRF analysis of glass samples.
And these are useful for several aspects, including the --
it helps to answer the question, when is a peak, a peak?
This topic got started
at an elemental analysis working group discussion
where the XRF users broke off and we were talking about some
of the issues that we face, including what do we do
with some, some of the small peaks that we run
across in glass cases?
I put arrows on several of these down here
and we were discussing, would you call that one a peak?
Would you use that in element ratio
when you're comparing glass samples?
We recognized there was a need for standardization
to make those decisions, and during that discussion,
we brought up the topic of the possibility of signal
to noise ratios to help those assessments.
And after that discussion, we decided to explore the signal
to noise ratios further and see where that led,
and that actually led into more than just is it a peak,
and should we use it ratios,
and that's what I'll be sharing today.
It's divided into three sections,
I've got some definitions of terms.
The reasons why you may want to calculate signal to noise ratios
when you're doing XRF on glass.
And also how to calculate those signal to noise ratios.
Just a quick word of warning before we get going
so you know what to expect, statisticians, don't worry,
you'll, you'll be fine here.
All right.
The first definition, first term is signal to noise ratio,
and when you're doing this, signal is you're trying
to measure how much analyte is present, the noise is anything
that interferes with that measurement of the signal,
random fluctuations from the environment
or from the instrument.
Signal is oftentimes defined as the peak area,
that's how you calculate that.
The noise in most typical applications is the standard
deviation of the baseline when the analyte is absent,
and the signal to noise ratio is simply signal divided by noise.
This doesn't, however, work for XRF analysis
because you're not able to take out the glass sample or take
out the strontium out of this example
to get what the noise would be, what the standard deviation
of the baseline would be because the background is dependent
on the matrix.
So, we need a way to calculate signal and noise
on the same spectrum, and people have done that.
Signal to noise ratio for XRF is just a little bit
different then.
The signal is still the peak area, that area in green.
And the noise then, is not true noise, but it's the square root
of the background counts under the peak of interest.
SNR is signal divided by this noise, and like I said,
this has been around for a while, it's 1975,
IUPAC nomenclature rules.
So it's an application of what's already been out there.
The next definition, limit of detection, or LOD.
The LOD is the lowest concentration of an analyte
that can be reliably detected, and it corresponds to a signal
to noise ratio for a peak of 3.
And at 3, a peak doesn't look very good,
there's about a 90% confidence level
that that analyte is actually present,
but there's enough information there that you're able
to say there is a peak.
This was taken from the 1975 rules and also supported
by a 1980 ACS paper in analytical chemistry.
Limit of quantitation, so now you've -- you have a peak there,
you've got it detected, but now you want to be able
to quantitate it, and the LOQ is the lowest concentration
of that analyte that could be reliably quantified,
corresponds to a signal to noise ratio of 10 for a peak.
And this is from that same 1980 paper.
So now I get into the reasons why you would calculate signal
to noise ratios when you're doing XRF analysis of glass.
Question that comes up, when is a peak a peak?
And if we apply those rules, the 1975 rules,
you can label a peak anytime the signal to noise ratio
of that peak is at least 3.
A caveat includes interferences, and these can be
like Kristine mentioned earlier, you assess your spectrum to find
out what's going on, and you can get interferences
from other elements, sum peaks, escape peaks, system peaks.
Here's an example of a glass case,
I think this was the second case I ran
after we modified our procedures to use this, and I had --
I'll expand to this region -- I had some elements,
some possible peaks, I wasn't comfortable necessarily just
labeling right out.
The titanium looked like it was there,
but it didn't look very good.
Copper and zinc kind of looked okay,
but not anything I would just jump
out and, and label right away.
Strontium zirconium, I just circled for comparison here.
Calculated that a signal to noise ratios for these peaks
and titanium was 6, copper and zinc were in the 5 range,
and so all three of these can be labeled on your spectrum.
Okay. Now you have it labeled and you want to know,
should I use it in a ratio?
This helps you to select the elements for ratio comparisons,
and to apply the 1980 American Chemical Society Guidelines
because we want to use reliably quantified values,
any peak with a signal to noise ratio of at least 10 can be used
in these element ratios
for doing semiquantitative comparisons
with the same caveats.
This one was the first case I performed
after we changed our procedures to look, look at it this way,
and I had several, several peaks that looked pretty decent,
but they seemed not very strong, I wasn't sure
if I should use those for comparisons for some
of semi quant comparisons.
I didn't know if I should do manganese iron as an example,
or rubidium iron, or strontium zirconium,
so I calculated the signal to noise ratios for each of these,
and for every one of them, the signal to noise ratio was
at least 10, so I was able to use all of those in ratios.
And there you can get a sense for what the peaks look
like at that LOQ value.
How low can you go?
What are the limits of detection
for your instrument using your method?
What you do is you estimate the limits of detection
from standard glasses,
these standard glasses are standard reference material
like 1831, it can be FGS, the German standard glasses
that have known concentrations, or reported concentrations
of several different elements in that glass sample.
And to find out the LOD, you use this equation,
3 times the reported concentration of that element
and that standard, divided by the signal to noise ratio
of that element's peak.
And I'll show you what that looks like on a spectrum.
Here I have SRM 1831 with a reported titanium concentration
of 114 parts per million,
and if you see the titanium peak right there, it has a signal
to noise ratio of 7.1, so I plug those into the equation,
and I come out with an LOD of titanium using this instrument
and this method for -- of 48 parts per million.
Compare that to zirconium as another example,
and I got a limit of detection of 5.9 parts per million.
So you can see it's not equal from one place to another,
one element to another across the spectrum.
This next slide shows how much they actually vary.
This is a table showing average limits of detection calculated
from three different glass standards, run on my instrument,
it's a 100 micron monocap system for 1200 live seconds.
And you see it ranges from 7,000 at sodium, down all the way
to single digits for strontium zirconium.
The primary factor for this improvement involves the
critical escape depth
of the emitted x-rays from these elements.
In other words, how far through the glass can those emitted
x-rays travel and still make it to the detector,
and the critical escape depth for sodium is just 6 microns,
whereas for strontium, it's up over 2,000 microns,
so you're sampling a much greater depth
for the heavier elements.
And if you compare these to other published values
that show LODs for XRF, these are greatly improved
because the standard method is you analyze thin films
at 100 live seconds, and instead, this is doing it
on casework type examples, and this is full thickness,
hopefully most of the time you get full thickness samples,
but even if it doesn't, you, you do get improved,
compared to the other standard methods.
The next thing you can do, once you have the signal
to noise ratios is compare XRF systems,
this can be for purchasing reasons,
it can be for troubleshooting,
it can be to improve your methodology.
And what you do is you run the standard glass samples,
calculate your LODs and you compare them
from one system to another.
This slide shows a comparison of three different configurations,
this is collecting from standard reference material,
1831 for 1200 live seconds,
the first is a 100 micron monocap system.
The middle one's a polycap.
And the third one's a 300 micron monocap.
And you can see that they all follow the same general trend
of LODs going from thousands down to single digits,
but they do differ from one system to another.
And if you notice the titanium and strontium are highlighted,
those I'll show the comparison on the actual spectra
in the next few slides.
Here are un-normalized spectra from those three systems,
1831 for 1200 live seconds,
you can see that the polycap has a lot more signal
that the other two, and the monocap
of 100 micron has significantly less.
If we expand to this region,
you see the same increased intensity,
also increased background through that region,
and to compare these systems, you look at their LODs,
here's titaniums with the signal to noise ratio
and LODs for those spectra.
And then for strontium, so it's easy to compare systems.
The next slide shows three very similar systems,
each are 100 micron monocap systems.
The only difference between these three is
that the green trace was collected
for 1800 live seconds instead of 1200.
And calculated the LODs, and for titanium strontium,
they're fairly similar from one system
to another, as you would expect.
A final reason to calculate signal
to noise ratios is for QA/QC checks.
A couple of different ways you can go about it,
initial instrumental or method validation,
you can set up a signal to noise ratio target
from a standard glass such as SRM 612, which has a variety
of elements added to it at about 50 parts per million,
and what is suggested here is make sure that your signal
to noise ratio for each of those elements is at least 10,
and that's to make sure that your instrument is able
to detect all these and is able to, is sufficient
for forensic casework and this is the level
that we want to be at.
A second check is your daily function verification check,
which Kristine mentioned earlier, the suggestion
on this slide is an LOD target of titanium of no greater
than 50 parts per million, so if you run your 1831
at the beginning of your run, all you need
to do is calculate the signal to noise ratio for titanium
and make sure it's at least 6.8, and that ensures
that for semiquantitative comparisons that you conduct,
it is reaching good limits of detection.
Here's the example of standard graphics material 612,
which has that variety of elements added,
and just a comparison, the top spectrum is --
was collected for 1200 live seconds, the bottom for 20,000,
and just by increasing the collection time, you greatly,
you can easily increase your signal to noise ratio.
And so, at the 1200, the signal to noise ratio was 8.9,
and so if I would have collected for 1500, or 2,000 live seconds,
I would have most likely hit the signal to noise ratio of 10,
and that's what I could have used as my starting,
my initial instrument method parameters.
So just to review, the usefulness of signal
to noise ratios for XRF helps you to decided on peak labels
and the decision of when
to use those elements in semi quant ratios.
You can calculate limits of detection
and compare those limits to other systems,
and for QA/QC checks, you meet LOD or SNR thresholds.
So those are some of the reasons why you would, you may want
to calculate signal to noise ratios,
here's how those ratios are calculated,
and just so you don't worry, I have a spreadsheet set
up where it's a cut and paste into that, and within a matter
of seconds you get all the signal noise ratios
for that spectrum, or for each element in that spectrum,
so it's not a tedious, hands-on process.
But here's what the software is doing.
Here's 1831, and I'm just going to use strontium
in as an example, so expand to that region.
The first thing you want
to do is identify the channels of interest.
Next you define the baseline
that separates the signal from the background.
And finally, you perform your calculations.
The total counts,
which hopefully your software can give you
through those channels, and then the background counts,
it's a fairly easy calculation to perform there.
The signal is the total minus the background
and the noise is the square root of the, those background counts,
so you're able to calculate the signals
to noise ratio for this peak is 36.
Just calculated it for other elements across the spectrum,
just as a demonstration, and that's how easy it was
to do that, I put it into the software and I got all of these
in a matter of seconds.
These are the references, the foundational works
that I used for this topic.
I'd like to acknowledge my employer,
the Michigan State Police for getting --
giving me the time to conduct this research and also to come
down here to present it, all of the EAWG members
who helped me quite a bit through this whole process
to figure out what I needed to be up here telling you guys,
and the NIJ grant that funded that effort.
Thank you for your time and attention.
[ Applause ]