Rapid and Accurate ED-XRF Quantification of Trace Arsenic in Rice-Based Foods Employing ANNs to Resolve Lead Spectral Interference

Trace quantifications of arsenic (As) in foods by energy-dispersive X-ray fluorescence (ED-XRF) spectrometry are hindered by spectral overlap from lead (Pb) at characteristic emission lines. This study employed artificial neural networks (ANN) to statistically model and correct for As/Pb spectral overlap, enabling accurate As quantifications in rice-based foods. Calibration standards were prepared by pelletizing milled rice spiked with As and Pb, and validation was performed using a certified reference material, commercial rice-based foods, and Pb-spiked commercial foods. As calibration metrics were great (R2 = 0.92, standard error in calibration = 41.20 µg kg−1). The validation assessment achieved acceptable error using the As reference material (−19.43% error) and in commercial rice-based foods containing low Pb content (6 of 11 As determinations in agreement with the reference method). Additionally, accurate predictions of As were found in the presence of significant Pb interference (absolute mean error = 14.11% in Pb-spiked commercial foods). Overall, ANN modeling for Pb exhibited poor performance during both calibration and validation. This work demonstrates the usability of an ANN to address the As/Pb overlapping issue while offering insights into the strengths and weaknesses of ANNs when coupled with ED-XRF for trace elemental quantifications in foods.