Bayesian approach to automatic mass-spectrum peak identification in atom probe tomography.

Abstract Identification of mass-spectrum peaks is an indispensable step of an atom-probe tomography reconstruction process and can be a time-consuming procedure, vulnerable to errors, if performed manually. We propose a Bayesian approach to the peak identification problem, based on ranking of candidate ions according to their calculated posterior probabilities. The sample model is reconstructed by iteratively accepting top-ranked ions while taking into account prior information, models of experimental errors, and the already accepted ions. The designed approach has been applied to a number of time-of-flight mass spectra, measured for inorganic samples, and enabled a reliable construction of sample models, consistent with the results of manual analysis. Additionally, a “sliding window” approach for an accurate and efficient peak decomposition of a mass spectrum was established on the base of Fisher information.