Abstract

Journal of Near Infrared Spectroscopy
Volume 16 Issue 3, Pages 151–157 (2008)
doi: 10.1255/jnirs.773

Self-modelling curve resolution of near infrared imaging data

Christelle Gendrin,^a,b,* Yves Roggo^a and Christophe Collet^b
aF. Hoffmann-La Roche AG, Grenzacherstrasse 124, CH-4070 Basel, Switzerland
^bLSIIT–UMR CNRS 7005 Pôle API, Strasbourg University, Boulevard S. Brant, F-67400 Illkirch, France

Over the last few years, there has been renewed interest in near infrared (NIR) systems thanks to the development of focal plan array detectors allowing acquisition of both spectral and spatial information in a short time. However, this analytical technique generates a large amount of data, and advanced multivariate methods have to be used in order to extract the information of interest. When no information about the sample is known, spectral signatures (pure spectra) and mixing coefficients (concentration) of the compounds have to be extracted from the observed spectra to localise and identify the chemical species. In analytical chemistry, self-modelling curve resolution (SMCR) methods aim to unravel pure compound information from the mixed spectra. In this study, the performances of several algorithms for the SMCR of NIR imaging data are compared. The first is principal component analysis (PCA) which has been widely used in the NIR spectroscopic community. The others are multivariate curve resolution-alternating least squares (MCR-ALS), non-negative matrix factorisation (NMF) and positive matrix factorisation (PMF). The last three algorithms apply non-negativity constraints on both mixing coefficients and spectral signatures to achieve the deconvolution of the mixed spectra. The results demonstrate that, due to the application of mathematical constraints, PCA extracts spectral information which is difficult to interpret and relate to a chemical compound. Among the alternative methods, PMF leads to accurate extraction results and a good match with the reference spectra but it is also an algorithm which requires accurate tuning. This study opens up alternative ways of finding distribution maps and pure spectra in NIR imaging data.

Keywords: NIR imaging, self-modelling curve resolution, principal component analysis, pharmaceuticals, tablet

---

Full-text article (872 kB) (subscribers only)

Buy article on-line for £20 (get immediate access)

Alerting Service

 RSS Feed

Permalink: http://dx.doi.org/10.1255/jnirs.773
QR Code (what is this?):

---