Control oriented sample characterization using systematic multivariate statistical method

The identification and decision-making for biomedical samples can be taken via Raman spectroscopic technology. Conventional approach for Raman spectral analysis is to search the peak spectrum location directly so as to determine relevant Raman shift information. In many cases it is hard to find specific Raman band spectrum that is suitable for discrimination among various samples. This is due to the influence from numerous types of slowly varying and rapidly varying noises. A proper pretreatment must be conducted prior to Raman data analysis. Multivariate statistical technique can then be employed to take advantage of simultaneous information presented in the whole spectrum or in any specific parts. Multivariate spectral analysis has already been implemented and reported for absorption or reflectance spectroscopy, while it is applicable to the analysis of Raman spectrum as well. Raman spectroscopy can indicate molecular composition and it enables the detection of sample pathological changes in a non-destructive manner. Due to its nature as a weak signal, Raman spectrum is undoubtedly sensitive to most types of noises. Wavelet technique is therefore used for noise filtering to an acceptable level. To eliminate the scatter effect due to varying conditions, standard normal variate (SNV) can be processed so that all spectra can contribute equally. A dimension reduction approach - principal component analysis (PCA) is then proposed for sample characterization to capture significant signatures. Eventually different principal components are selected for clustering and various samples from different organs can be differentiated. A feasible approach for sample analysis is formulated and satisfactory results are obtained