Authors:
Valeriu Vrabie
1
;
Cyril Gobinet
2
;
Michel Herbin
1
and
Michel Manfait
2
Affiliations:
1
CReSTIC, Université de Reims Champagne-Ardenne, France
;
2
MéDIAN, CNRS UMR 6142, Université de Reims Champagne-Ardenne, France
Keyword(s):
Raman Spectroscopy, Paraffin-Embedded Cutaneous Biopsies, Blind Source Separation, Independent Component Analysis, Non-negative Matrix Factorization, Maximum Likelihood Positive Source Separation.
Related
Ontology
Subjects/Areas/Topics:
Applications and Services
;
Biomedical Engineering
;
Biomedical Signal Processing
;
Computer Vision, Visualization and Computer Graphics
;
Medical Image Detection, Acquisition, Analysis and Processing
Abstract:
Raman spectroscopy is a powerful tool for the study of molecular composition of biological samples. Digital processing techniques are needed to separate the wealthy but complex information recorded by Raman spectra. Blind source separation methods can be used to efficiently extract the spectra of chemical constituents. We propose in this study to analyze the performances of four blind source separation methods. Two Independent Component Analysis methods using the JADE and FastICA algorithms are based uniquely on the independence of the spectra. The Non-Negative Matrix Factorization takes into account only the positivity of underlying spectra and mixing coefficients. The Maximum Likelihood Positive Source Separation assumes both the independence and positivity of the spectra. A realistic simulated dataset allows a quantitative study of these methods while a real dataset recorded on a paraffin-embedded skin biopsy provides a qualitative study.