Neural network classification of pharmaceutical active ingredient from near infrared spectra

Abstract

This paper presents results from a scoping study undertaken with the intention of demonstrating the applicability of reflectance Near-Infrared (NIR) spectroscopy in classification of pharmaceutical type based on active ingredient. The part of the process selected for the scoping study is the packaging step in the manufacturing of pharmaceuticals. The rationale for the selection of the packaging step is in older processing lines the product is classified after tablet coating and before blister packaging using visual automatic inspection techniques however for 100% conformance a more discriminating technique is required.

In this study, NIR spectra (with wavelengths between 400nm and 1100nm) were obtained for samples pertaining to 3 different types of pharmaceuticals Quetiapine, Ibuprofen and Paracetamol. The dimensionality of the data set was reduced using Principal Components and the data was feed into a back propagation neural network configured to classify the data based on active ingredient type. The recognition rates achieved in this study were high enough to suggest that NIR spectroscopy is a viable method of ensuring 100% identification of pharmaceutical type prior to packaging for the pharmaceuticals tested.