Sia Afshari, Program Manager XRF, EDAX
As I am new at EDAX, my colleague Laurie Krupa sent me a copy of the August 2013 EDS Technical Note on “Mapping Termination Criterion” by Patrick Camus after our conversation about forensic applications.
As I read the technical note, one of those “Déjà Vu AOA” moments took me back to another application a long time ago.
In my previous job, I was involved with the development of a portable XRF device for a determination of lead in paint application. The EPA/HUD Guidelines were cited and prescribed faithfully as the Operating Protocols (OP) due to the litigious nature of this application and any deviation from these protocols were forbidden!
The HUD Guidelines, at that time, were mandating an average of three 30-second measurements at each location without any reference to the expected error and the degree of confidence that were required for quantification of a measured value. Digging in the archives for the origin of this OP, it became clear that these outdated protocols were drafted from the User’s Manual recommendations of the original portable XRF system that was manufactured based on 1970’s technology.
Mandating a fixed-time-based measurement by the EPA/HUD without inclusion of the measurement’s statistical error and the degree of confidence led to the collection of meaningless data that were eventually challenged and repeated at a significant cost. Based on those protocols, there was no distinction in the two measurements below as long as they were performed in 30 seconds:
1.0 ± 0.1 mg/cm2 and 1.0 ± 1.0 mg/cm2
Additionally, the fixed time protocols prevented the effective utilization of modern XRF systems that could perform the intended measurements in a much shorter time and with a higher degree of accuracy based on a set of published performance characteristics that defined the expected error and statistical confidence for each measurement as a function of time and substrate.
It took determined pressing of the issue for a couple of years and a change in the administration to replace the status quo with scientific reasoning to publish a set of Performance Characteristic Sheets as a part of what are the EPA/HUD Chapter 7 Guidelines. New protocols were based on statistical modeling of the data that were obtained from analyses of real-life samples and under normal operating conditions.
Back from the trip down memory lane, in my view, a similar situation seems to be present in the spectral imaging data acquisition for forensic application. The existing OP requires a fixed termination time without reference to the statistical significance of the count rate per pixel that defines the quality of the image. The input x-ray count rate to generate an image is a unique characteristic of the instrument used and can vary drastically from one system to the next.
In the referenced technical note, Pat Camus skillfully presents and concludes that “Spectral imaging data sets should be acquired with a termination criterion that provides a statistical level suitable for correct interpretation of the data”. The enclosed images in his technical note clearly support his position and the necessity for including the statistical expectation as a part of the measurement criteria!
The existing spectral imaging protocols are probably drawn from an older generation of XRF operation procedures where extended measurement time for acquisition of an image was necessary due to the low count rates. The modern XRF systems with high intensity micro-focus x-ray tubes, polycapillary x-ray optics, Silicon Drift Detectors (SDD), and advanced software algorithms can generate and process much higher count rates than their earlier generation, resulting in consistent, repeatable, and higher quality images for forensic application in a much shorter time.
As with the EPA/HUD experience, a new scientific approach should be considered in imaging protocols that is based on spectral count per pixel rather than a fixed measurement time. Implementing a statistical approach in data acquisition will utilize the technological advances in instrumentation, increase the optimization of analyst and instrumentation time, and provide more consistent results in forensic application for direct comparison of data sets.
After all, statistics do matter!