Automatic Baseline Subtraction of Vibrational Spectra Using Minima Identification and Discrimination via Adaptive, Least-Squares Thresholding

Abstract

A method of automated baseline correction has been developed and applied to Raman spectra with a low signal-to-noise ratio and surface-enhanced infrared absorption (SEIRA) spectra with bipolar bands. Baseline correction is initiated by dividing the raw spectrum into equally spaced segments in which regional minima are located. Following identification, the minima are used to generate an intermediate second-derivative spectrum where points are assigned as baseline if they reside within a locally defined threshold region. The threshold region is similar to a confidence interval encountered in statistics. To restrain baseline and band point discrimination to the local level, the calculation of the confidence region employs only a predefined number of already-accepted baseline minima as part of the sample set. Statistically based threshold criteria allow the procedure to make an unbiased assessment of baseline points regardless of the behavior of vibrational bands. Furthermore, the threshold region is adaptive in that it is further modified to consider abrupt changes in baseline. The present procedure is model-free insofar as it makes no assumption about the precise nature of the perturbing baseline nor requires treatment of spectra prior to execution.

Development of a classification model for non‐alcoholic steatohepatitis (NASH) using confocal Raman micro‐spectroscopy

“…Spectrum processing was then performed using MATLAB (Mathworks, Massachusetts) where the narrow spikes induced by cosmic rays were first removed. Then, as the Raman spectra of tissues typically contains Raman scattering, intrinsic tissue fluorescence and noise, background fluorescence from the tissue was estimated and removed using the adaptive minimax method [22]. Based on the method developed by Cao et al [23], polynomial fits [24] (based on the fluorescence-to-signal (F/S) ratio) were used to minimize the residual mean square (RMS) error.…”

Investigating the Responses of Cronobacter sakazakii to Garlic-Drived Organosulfur Compounds: a Systematic Study of Pathogenic-Bacterium Injury by Use of High-Throughput Whole-Transcriptome Sequencing and Confocal Micro-Raman Spectroscopy

“…The preprocessing of the raw Raman spectra can separate and subsequently eliminate the side effects which may influence the quality of spectroscope-based chemometric models and multivariate analyses. We first conducted a polynomial background fit (45) combined with baseline subtraction using identification and discrimination of minima via adaptive and least-squares thresholding (46) to remove fluorescence background derived from C. sakazakii cells on gold-coated microarray slides, Gaussian noise, white noise, CCD background noise, and cosmic spikes (47)(48)(49). Besides fluorescence background, most of the spectral interference is contributed by CCD background noise, which is generated due to the thermal fluctuations on the CCD detector.…”

Automatic Baseline Recognition for the Correction of Large Sets of Spectra Using Continuous Wavelet Transform and Iterative Fitting

“…Various authors tackled this issue by proposing automated or semi-automated algorithms for baseline correction that reduce human intervention. Effective solutions were devised using iterative polynomial fitting, 13 penalized quantile spline regression, 14 adaptive least squares/ Whittaker smoother, [15][16][17] moving average-peak stripping, [18][19][20] local second derivative, 12 and morphological or geometrical approaches. 21,22 The performance of these methods differ in terms of accuracy, computational speed, amount of human intervention, and types of spectra to which they can be applied; these goals are usually conflicting.…”

“…This even caused some authors to introduce user-defined knowledge back into algorithms that were initially devised as parameter free, e.g., to distinguish broad spectral bands from baseline sections with a high curvature. 15,23 In our opinion, getting completely rid of instance-related background knowledge may not even be achievable, not least because the very definition of what constitutes ''true'' signal and noise depends on the particular investigation.…”