Multiscale peak identification for bacterial SERS spectra

A multiscale approach for peak identification and representation is proposed in the context of surface-enhanced Raman spectroscopy (SERS). The issue of overlapping effects, which frequently appear in biological signals but have not been addressed well in established approaches, is the main concern in this paper. The peak identification will be an ambiguous problem if the overlapping effects are not addressed well. We propose a refinement on our multiscale peak identification (MSP) approach to apply matching pursuit for overlapping effects. The spectral peaks are modeled by the pattern matching method on their preferred scales estimated with the evolution of wavelet transform modulus maxima (WTMM). With the MSP approach, we are able to reconstruct a given SERS spectrum into a set of underlying Gaussian peaks. In the demonstrated experiment regarding the life cycle of Escherichia coli, more meaningful but not steadily accessible features can be derived. Moreover, the explicit peak representation can make the association of SERS spectrum with specific molecular information possible.