Is it really Quant mapping if we don't collect standards?

[kthompson75]

Quant mapping - in either %wt or %at - is now very standard in EDS mapping. The advantages of background subtraction and peak deconvolution are readily apparent. However, the final step in assigning a quantitative % to each pixel can be incredibly misleading. Here are two traps that the vast majority of EDS users fall into:

Trap #1. No standards are collected. For aesthetics purposes, the data is normalized to 100% in each pixel. Hmmmmm. Do enough standards-based quant and we realize that standardless quant just doesn't hold up to the same level of accuracy as when standards are collected. Why pretend that maps are ideal representations of quantitative analysis.

Solution #1. The Noran System 7 fully supports user-collected standards during mapping. Quant maps are only truly Quant maps when stadards have been collected.

Trap #2. Most EDS mapping only collects between 100 - 1000 x-ray counts per pixel. Straight counting statistics says that 100 counts per pixels provides a 1 sigma standard deviation of +/-10%. This means that the answer on pure element standards is confidently measured (depending on your threshold for confidence interval) between 70% - 130% based purely by counting statistics.

Solution #2. Collect more x-rays or run an algorithm that interpolates neighboring pixels to improve the counting statistics.

Need further proof? Collect a map of Net counts, k-ratios or standards-based quant on a pure element standard. Export to excel and measure the standard deviation. A pure element standard should always be 100% and purely homogenous. How close is the map?

Edit by John: This will amuse you: recently now that we have a full x-ray mapping quant, people are no longer looking only at raw data and so the question I sometimes get is this. "The totals on my x-ray scans range from 90 to 110%. Why is that?"

So I'll ask: "What dwell time per pixel are you using?" and they'll often answer: "Oh, 20 msec..." and so I'll respond: "Better try 200 msec per pixel!"

[John Donovan]

Good stuff, Keith.

I will simply point out that NIST suggests only quoting "major", "minor" or "trace" instead of numerical values when using standardless methods.

Did you you see my previous standardless EDS "rant" here?

http://probesoftware.com/smf/index.php?topic=302.msg1530#msg1530