Premium
Comparative study of multivariative analysis methods of blood Raman spectra classification
Author(s) -
Bratchenko Lyudmila A.,
Bratchenko Ivan A.,
Lykina Anastasiya A.,
Komarova Marina V.,
Artemyev Dmitry N.,
Myakinin Oleg O.,
Moryatov Alexander A.,
Davydkin Igor L.,
Kozlov Sergey V.,
Zakharov Valery P.
Publication year - 2020
Publication title -
journal of raman spectroscopy
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.748
H-Index - 110
eISSN - 1097-4555
pISSN - 0377-0486
DOI - 10.1002/jrs.5762
Subject(s) - linear discriminant analysis , pattern recognition (psychology) , principal component analysis , chemometrics , artificial intelligence , multivariate statistics , raman spectroscopy , sensitivity (control systems) , logistic regression , multivariate analysis , statistics , mathematics , chemistry , computer science , chromatography , optics , physics , electronic engineering , engineering
The pathological state of a human body leads to altered biochemical composition of body fluids. Conventional biochemical analysis of body fluids is notable for its low‐informative value in localizing a particular pathology. As an alternative, Raman spectroscopy provides detailed evaluation of blood characteristics at the molecular level. Raman blood spectra are characterized by multicollinearity as well as by the presence of autofluorescence background and noises of different nature. Choice of a proper method for experimental data processing of blood spectra is crucial for obtaining statistically reliable information regarding a pathological process in the body. This study examines different approaches to multidimensional analysis of the various‐size Raman spectral dataset of human blood samples by a cost‐effective Raman setup in a clinical setting. To discriminate blood samples by the pathology type, statistical processing of experimental data is performed by factor analysis , logistic regression , discriminant analysis , classification tree , projection to latent structures discriminant analysis ( PLS‐DA ), and soft independent modeling of class analogies . Comparative analysis of the discussed multivariate methods demonstrates that the PLS‐DA method (sensitivity 0.75, specificity 0.81, and accuracy 0.76) proved to be the most effective for the classification of blood samples by cancer localization. In terms of classification for the presence of hyperproteinemia, the most efficient are the logistic regression method (sensitivity 0.89, specificity 0.99, and accuracy 0.97) and the discriminant analysis method (sensitivity 0.83, specificity 1.0, and accuracy 0.97). In general, the selected multivariate methods could serve as a reliable tool for analyzing spectral characteristics of body fluids.