Integrating hyperspectrograms with class modeling techniques for the construction of an effective expert system: Quality control of pharmaceutical tablets based on near-infrared hyperspectral imaging.
Journal:
Journal of pharmaceutical and biomedical analysis
PMID:
39881455
Abstract
Near-infrared hyperspectral imaging (NIR-HSI) integrated with expert systems can support the monitoring of active pharmaceutical ingredients (APIs) and provide effective quality control of tablet formulations. However, existing quality control methods usually test a limited number of variability sources affecting the final product. This study examines the potential of NIR-HSI (in the spectral range of 935.61-1720.2 nm) as an advanced and high-throughput detector to identify different manufacturing factors and their fluctuations that impact tablet properties. These are, for instance, particle sizes of powdered excipients, their mixing, compression force used to form a tablet, origin of ingredients, storage conditions, and concentration of API. During the study, the novel expert system approach was developed to support NIR-HSI, enabling the detection of subtle, diverse substandard anomalies in tablets. The system combines (i) hyperspectrograms, which characterize and simplify tablet spatial heterogeneity through principal component analysis scores distribution, and (ii) a one-class classifier (OCC), trained exclusively on target class samples, without the need for substandard tablets. The system was trained to recognize known sources of variation and validated using tablets with cellulose, magnesium stearate, and ascorbic acid as API. It outperformed the alternative approach based on averaged spectra, achieving 100.00 % sensitivity and 98.77 % specificity.