Utilization of spectral vector properties in multivariate chemometrics analysis of hyperspectral infrared imaging data for cellular studies.

A suite of numerical techniques was utilized in a concerted fashion for the efficacious multivariate chemometrics analysis of hyperspectral infrared imaging data of exfoliated oral mucosa cells. Based on the vector representation of infrared spectrum a1xnu), spectral vector properties (SVP) are demonstrated to possess underpinning spectral information that was exploited in crucial chemometrics analyses; which include outlier spectra identification, selection for a subset of imaged mid-infrared spectra that contain good oral mucosa cell signals, and, for the first time, obtain major biochemical constituent spectra via the band-target entropy minimization (BTEM) curve resolution algorithm. The relative concentration spatial distribution of the major biochemical constituents observed, namely membrane lipids and various cellular protein structures (alpha-helix, beta-sheet, turns and bends), were subsequently acquired through multi-linear regression and were displayed as chemical contour maps. Amongst the set of numerical algorithms employed, two novel unsupervised clustering algorithms were developed and tested. One is useful for outlier spectra detection, and the other aids the selection of pertinent spatially distributed spectra that possess oral mucosa cell mid-infrared spectra with good signal-to-noise ratio. It is anticipated that this developed numerical suite will serve as an effective multivariate chemometrics protocol for cellular studies and biomedical diagnostics via infrared imaging.