Effects of Baseline Correction Strategy on Cervical Raman Spectral Analysis in a Late-Gestation Rat Model

Fluorescence background is a persistent limitation in quantitative Raman spectroscopy of biological tissue, and it is pronounced in the pregnant cervix, where weak Raman bands sit on a strong and variable baseline. Baseline correction is a routine step, yet its effect on the final result is rarely tested in a controlled way. We examined how the choice of correction strategy affects spectral metrics and the biological difference inferred between two gestational stages. Ex vivo cervical spectra from a rat model at gestational days 15 and 20 were processed with fixed-degree polynomial fitting from the second to seventh degrees, five-point piecewise linear correction, an automated polynomial routine, and asymmetric least squares. Correction quality was assessed by the ratio of Raman-active to -inactive regions and by whether each method preserved the direction of gestational change. The choice of method altered the quantitative output and, for some methods, reversed the inferred direction, because the automated high degree of fit over-suppressed the spectrum and inflated the metric. Moderate-degree polynomial fitting and asymmetric least squares gave the most coherent and biologically consistent results. Baseline correction therefore determines cervical Raman outputs and should be selected and reported transparently.