34168

Comparative Study of Different Multi-way Models and Preprocessing Algorithms for the Analysis of Ointments using Raman Chemical Imaging

Favoritar este trabalho

Raman chemical imaging (Raman-CI), as an evolving chemical imaging technique, becomes of major interest to pharmaceutical manufacturer. Raman-CI is used as a tool for enhancing drug quality and controlling manufacturing processes. In Raman-CI sample measurement generates a hyperspectral data cube that exhibits Raman spectrum for every pixel in the image. As such; preprocessing algorithm, to neutralize the effect of noise; as well as different multi-way data analysis models are of great importance to Raman-CI analysis. The aim of this study was to compare the prediction ability of different multi-way regression methods. Three two-way regression models and three three-way models are compared. The two-way models discussed in this study are; Classical Least Squares Regression (CLS); Partial Least Squares Regression (PLS); and Principal Component Regression (PCR). While, the three-way regression models discussed in this paper are; Parallel Factor Analysis (PARAFAC); Tucker3; and Tucker2 models. The algorithm behind each of these models is discussed in details, with step-by-step guidance for computing the regression coefficient matrix as well as interpreting different diagnostic graphs. Each of these models was evaluated using four different preprocessing methods. The study is carried out using concentration gradient of an API in ointment base. The hyperspectral images were captured using Raman Chemical Imaging instrument. The study shows that the two-way models appear to have superior performance over multi-way models, where PLS and CLS showed significantly higher prediction ability than other models.