The research article explores the use of Hartigan-Wong k-means cluster analysis to classify the Raman spectra of thousands of particles aerosolized from an Advair Diskus® 100/50 inhaler. This method is compared to the result of the manual assignment method, which can be time-consuming and vary between instrument operators.
The study aims to develop an automated classification method that could save time during the analysis of Morphologically Directed Raman Spectroscopy (MDRS) data and easily be applied to new materials and products. The method adds to a growing suite of microstructural characterization tools to aid in product development and regulation of inhaled pharmaceutical products.
Particles from an Advair Diskus 100/50 inhaler were collected using a previously reported bespoke aerosol collection apparatus. The particles were imaged, their positions automatically tracked using white light thresholding, and the individual Raman spectra of at least 3000 particles were collected. The particle spectra were then clustered using the Hartigan-Wong k-means clustering algorithm.
The results of the cluster analysis were visualized, and the fraction of particles by number in each agglomerate class from three repeat MDRS measurements analyzed using both manually assigned classification rules and k-means cluster analysis are shown. The most striking difference between the two plots is the reduction in apparent run-to-run variability in the cluster analyzed data compared to that using manually assigned classification rules.
