Developing analytical methods for identification of Elongate Mineral Particle (EMP): investigating the possibility of using FTIR spectra and machine learning modeling classification tasks in asbestos containing samples

Abstract

Background: EMPs pose substantial public health risks due to their well-documented associations with respiratory diseases, including asbestosis, lung cancer, and mesothelioma. Current asbestos identification methods, such as polarized light microscopy (PLM) and transmission electron microscopy (TEM), have limitations related to subjectivity, cost, and complexity. Fourier Transform Infrared Spectroscopy (FTIR) offers a promising alternative, but its application in EMP classification remains underexplored. This dissertation investigates the potential of using FTIR in combination with Partial Least Squares Discriminant Analysis (PLS-DA) for improving EMPs detection. Objectives This research aims to (1) develop an analytical method utilizing FTIR-DRIFTS and PLSDA to classify EMPs accurately, (2) investigate and validate the PLS-DA model performance of this method in laboratory-controlled conditions, and (3) improve asbestos monitoring, reduce occupational exposures, and support regulatory compliance in public health and industrial hygiene settings. Methods: Six regulated asbestos reference materials were analyzed using FTIR-DRIFTS. Spectral data were preprocessed and used to train PLS-DA models for classification. The models were validated against reference standards, potential interference, and bulk samples with different mineralogical compositions. The study also assessed the effects of background variability, particle size, and spectral interference on classification accuracy. Results: The developed FTIR-PLS-DA method demonstrated high classification accuracy, correctly identifying asbestos types with minimal misclassification. The technique effectively distinguished EMPs based on their vibrational spectra, with model detection ability reaching as low as 0.0016% by weight for some asbestos classes. Background variability was found to impact model performance, emphasizing the importance of same-day spectral calibration. The model was robust across different particle sizes and sources, though slight misclassification occurred between closely related amphibole types. Conclusion: FTIR-PLS-DA can provide a reliable, reproducible, and cost-effective approach for asbestos and EMP identification. The method minimizes analyst subjectivity and enhances classification accuracy, making it a valuable tool for occupational and environmental monitoring. With a potential integration of handheld IR spectrometers that could further advance on-site asbestos screening, reducing health risks for workers. Public Health Impact: By improving the selectivity and accuracy of EMPs detection, such a method can enhance workplace safety, facilitate regulatory compliance, and support early intervention in asbestos related exposure risks.