Inter-equipment ongoing validation of microplastic identification by micro-FTIR using optimal resources

FTIR is widely used for identifying microplastics due to its performance and affordable equipment. However, there is a need for a universal method for the automatic identification of microplastics by FTIR, robust to different spectra collection conditions, and capable of adapting to the diversity of particle spectra. This research presents a method for developing a reliable and simple algorithm for inter-instrumental microplastic identification using micro-FTIR. Spectra collected from different equipments and settings across various laboratories, from microparticles previously confirmed as PET (positive cases) and non-PET (negative cases), were compared with a PET reference using multiple algorithms (match methods) based on six weighted and unweighted correlation coefficients. After excluding low signal-to-noise spectra with a newly developed universal algorithm, the 5th percentile of the match values for positive cases, estimated using a robust bootstrap method, was used as a minimum match (P5»P) to ensure a 95% true positive rate (TP). Assuming a normal distribution of the match values for negative cases, the false positive rate (FP) was calculated. The best method identified uses an unweighted correlation of differentiated signals, resulting in a P5»P = 0.4140 and FP = 0.0005% for spectra with a minimum signal-to-noise ratio of three. This identification performance, estimated from 117 spectra, is statistically equivalent to that observed from the identification of 405 additional particles using the validated method, at a 99% confidence level set by the Clopper–Pearson method, thereby extending the method's applicability to 522 particles. The developed method for PET identification by FTIR is robust across various acquisition conditions and instruments, was implemented in a user-friendly MS Excel spreadsheet, and can be easily tested with more particles and different instrumental settings. The validity of the identification algorithm and the defined criteria cannot be extrapolated to a different reality with respect to spectral collection conditions and the diversity of negative cases without a prior performance assessment.