Background and Objective: The potential of using various multivariate analy
sis methods for classification of fluorescence spectra acquired in vivo fro
m laryngeal tissues in Patients was investigated.
Study Design/Materials and Methods: Autofluorescence spectra were measured
on 29 normal tissue sites and 25 laryngeal lesions using 337-nm excitation.
Four different multivariate analysis schemes were applied. Laryngeal fluor
escence spectra from patients who had been administered F-aminolevulinic ac
id (ALA) were obtained using 405-nm excitation and were classified using pa
rtial least squares discriminant analysis (PLS-DA).
Results: For autofluorescence spectra, logistic regression based on princip
al component analysis (PCA) or PLS, or PLS-DA all resulted in sensitivities
and specificities around 90% for lesion vs. normal. Using ALA and 405-nm e
xcitation gave a sensitivity of 100% and a specificity of 69%.
Conclusion: Multivariate analysis of fluorescence spectra could allow class
ification of laryngeal lesions in vivo with high sensitivity and specificit
y. PLS performs at least as well as PCA, and PLS-DA performs as well as log
istic regression techniques on these data. (C) 2001 Wiley-Liss, Inc.