Spectral feature extraction and characterization via simple linear and nonlinear technology integration

A systematic approach for spectral calibration, extraction and characterization is presented. It involves the simple technology integration of the linear and nonlinear methodologies. Raman spectra are essentially weak signals whose features are sensitive to a variety of noises, while the background spectra (e.g., fluorescence spectra) act as another primary intensity component in the raw Raman spectrograph. Thus the calibrated Raman spectra from measurements are subject to detrend and denoising so as to extract the intrinsic spectra, which contain solely the unique spectral signatures of the individual biomedical samples. By removing the slowly varying baseline background spectra and rapidly varying noises, intrinsic Raman spectra could be extracted using a combination of linear and nonlinear methodologies. The goal is to capture the distinguishable wave number information on the Raman shift for feature identification. This approach is crucial for spectra analysis to associate Raman spectra with the medical diagnosis. Considering the slowly varying nature of background spectra such as fluorescence, linear least squares estimation is applied to detrend the background spectra from the measured spectra being calibrated. Since most noises are rapidly varying, nonlinear discrete wavelet transform (DWT) denoising is applied to separate the intrinsic Raman spectra from noises for spectral identification. This approach is proposed so as to differentiate the normal spectra from the abnormal spectra of the fresh mice lung samples. Numerical simulation outcomes include the calibrated Raman spectra, detrended spectra free of background effects and the extracted intrinsic spectra after nonlinear DWT denoising. The satisfactory results indicate that the proposed simple linear and nonlinear technology integration provides a useful approach for biomedical sample characterization. In addition, it reduces computational complexity compared with artificial intelligence approaches, whi- h has no technical difficulty in real time medical diagnosis implementation. This serves as a fundamental step for further data clustering and accurate decision making among numerous diverse samples.