A comparation of different strategies in Principle Component Analysis (PCA) algorithm for clustering human tooth surface using Laser-Induced Breakdown Spectroscopy (LIBS)

Abstract

The aim of this work was to observe homogeneity of human tooth surface using classification technique by laser-induced breakdown spectroscopy (LIBS) coupled with principle component analysis (PCA) algorithm. The human tooth was irradiated by 110 mJ Nd-YaG laser (1064 nm) under Helium gas with flow rate of 50 ml/s to produce plasma. Photon emission of the plasma was captured by ocean optic spectrometer HR 2500+ and displayed spectra of intensity as a function of wavelength. The spectra data were analysed by different strategies in PCA algorithm for classifying human tooth surface. The spectra data were split into three ranges that were a full spectral window, FW (200-850 nm), long special spectral window, LSW (380 - 660 nm) and short special spectral window, SSW (550 - 600 nm). These selected suitable input variables using spectral windows can reduce the influence of over fitting phenomena on classification results. Prior to PCA analysing, data were treated by different strategies of pre-processing namely linear baseline correction, area normalisation, and no pre-processing. The results showed that the short special spectral window (SSW) using pre-processing of area normalization could either clustering and distinguishing parts of human tooth surface clearly. Conclusion dentin surface has highest homogeneity of all.