Simultaneous Quantification of Carbamazepine Crystal Forms in Ternary Mixtures (I, III, and IV) by Diffuse Reflectance FTIR Spectroscopy (DRIFTS) and Multivariate Calibration

Diffuse reflectance FTIR spectroscopy (DRIFTS) coupled with modern multivariate calibration methods, namely artificial neural networks (ANNs) in two versions (ANN-raw and ANN-pca), support vector machines (SVMs), lazy learning (LL) and partial least squares (PLS) regression, is used in this study fo...

Full description

Saved in:
Bibliographic Details
Published in:Journal of pharmaceutical sciences 2006-11, Vol.95 (11), p.2419-2431
Main Authors: Kipouros, K., Kachrimanis, K., Nikolakakis, I., Tserki, V., Malamataris, S.
Format: Article
Language:English
Subjects:
Algorithms
Biological and medical sciences
Calibration
carbamazepine
Carbamazepine - analysis
Crystallization
diffuse reflectance
DRIFTS
General pharmacology
lazy learning
Medical sciences
Microscopy, Electron, Scanning
Models, Statistical
Multivariate Analysis
neural networks
partial least squares
Particle Size
Pharmaceutical technology. Pharmaceutical industry
Pharmacology. Drug treatments
Powders
Spectroscopy, Fourier Transform Infrared
support vector machines
X-Ray Diffraction
Citations: Items that this one cites
Items that cite this one
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Diffuse reflectance FTIR spectroscopy (DRIFTS) coupled with modern multivariate calibration methods, namely artificial neural networks (ANNs) in two versions (ANN-raw and ANN-pca), support vector machines (SVMs), lazy learning (LL) and partial least squares (PLS) regression, is used in this study for the quantification of carbamazepine crystal forms in ternary powder mixtures (I, III, and IV). Two spectral regions (675–1180 and 3400-3600/cm) were selected and the data were partitioned into training and test subsets applying the Kennard-Stone design. It was found that all the selected algorithms perform better than the PLS regression (root mean squared error of prediction (RMSEP) from 3.0% to 8.2%). ANN-raw, trained on uncompressed spectral data, shows best predictive performance (RMSEP 
ISSN:0022-3549
1520-6017
DOI:10.1002/jps.20690