Signal-to-noise ratios in forensic glass analysis by micro X-ray fluorescence spectrometry T. Ernst, a * T. Berman, b J. Buscaglia, c T. Eckert-Lumsdon, d C. Hanlon, e K. Olsson, f C. Palenik, g S. Ryland, b T. Trejos, h M. Valadez i and J. R. Almirall h Micro X-ray fluorescence (m-XRF) spectrometry using an energy dispersive X-ray (EDS) detector is capable of detecting certain major, minor, and trace elements that permit potential discrimination of glass fragments in forensic cases on the basis of differences in elemental composition. Often, elements used for discrimination are present at concentrations near the detection limit of the EDS system, and the decision whether to utilize these minor peaks in a comparative analysis has generally been left to the discretion of the examiner. The use of signal-to-noise ratios (SNRs) of spectral peaks provides additional objectivity in peak identification/label decisions and in the selection of elements in semiquantitative ratio compar- isons. In addition, the use of SNRs enables calculations of limits of detection and limits of quantitation and the monitoring of instrument performance, and facilitates performance comparisons of different m-XRF configurations. This paper demonstrates a practical method for applying the concepts of SNR, limits of detection, and limits of quantitation to m-XRF generated EDS-based spectra, discusses the implications of such determinations, addresses spectral features that must be considered when making the calculations, and illustrates the application of these concepts to the example of forensic examination and comparison of glass samples. Copyright © 2012 John Wiley & Sons, Ltd. Introduction Glass samples are submitted to a forensic laboratory in a variety of investigative circumstances in which a glass object is broken, such as motor vehicle hit-and-run incidents, forced entries, and assaults. In these situations, broken glass may have transferred from the source to another object or person. An objective of a forensic glass examination is to compare samples to determine if they can be discriminated using physical, optical, and/or chemical properties (e.g., color, thickness, refractive index, density, and elemental composition). If the samples are distinguishable in any of these properties, it can be concluded that they did not originate from the same glass object. If the samples are indistin- guishable in all of these properties, the possibility exists that they originated from the same glass object. [1] The glass samples received for forensic examinations range from full-thickness fragments that are several centimeters in diameter to very thin particles that are less than 100 mm in diameter. The samples may have manufactured surfaces (flat or curved) or fractured, uneven surfaces. The geometry, size, and thickness of the submitted glass samples influence the examinations performed on the samples, the data generated from the samples, and the significance of the results. The use of an elemental analysis method such as micro X-ray fluorescence (m-XRF) spectrometry enables a high level of discrimination among different glass objects. [2,3] Interpretation of elemental data from glass samples often involves a combina- tion of spectral comparisons and peak intensity ratio compari- sons. [2–4] Spectral comparisons contrast the overall elemental profile of two samples, including the elements detected, the peak shapes, and the relative peak heights. Peak intensity ratios allow semiquantitative comparisons of elemental composition. Several forensic publications demonstrate the effectiveness of discriminating glass samples on the basis of one or both of these methods. [2–5] However, within these publications, there is little discussion in considering when a specific element is sufficiently above background for use in a forensic comparison. This paper presents applications of the long-standing principles of signal- to-noise ratio (SNR), limit of detection (LOD), and limit of quanti- tation (LOQ) to forensic glass examination by m-XRF for making analytical decisions more objective and standardized. The scientific basis for defining SNR, LOD, and LOQ has been well established in the literature. [6,7] Algorithms exist in some energy dispersive X-ray (EDS) software packages that utilize a form of these calculations; however, such calculations are * Correspondence to: T. Ernst, Michigan State Police Grand Rapids Laboratory, 720 Fuller Ave NE, Grand Rapids, MI 49503, USA. E-mail: ernstt@michigan.gov a Trace Evidence Unit, Michigan State Police Grand Rapids Laboratory, Grand Rapids, MI, USA b Orlando Regional Operations Center, Florida Department of Law Enforcement, Orlando, FL, USA c Counterterrorism and Forensic Science Research Unit, Federal Bureau of Investigation Laboratory, Quantico, VA, USA d Trace Evidence Branch, U.S. Army Criminal Investigation Laboratory, Forest Park, GA, USA e Miami-Dade Police Department, Miami, FL, USA f Johnson County Crime Laboratory, Olathe, KS, USA g Microtrace LLC, Elgin, IL, USA h Department of Chemistry and Biochemistry and International Forensic Research Institute, Florida International University, Miami, FL, USA i Crime Laboratory, Texas Department of Public Safety, Austin, TX, USA X-Ray Spectrom. 2014, 43, 13–21 Copyright © 2012 John Wiley & Sons, Ltd. Research article Received: 31 July 2012 Accepted: 4 November 2012 Published online in Wiley Online Library: 21 December 2012 (wileyonlinelibrary.com) DOI 10.1002/xrs.2437 13