Deterministic baseline correction and line-shape modeling for N₂O breath spectra from healthy and gastric-disease...

In this study, baseline correction and line-shape model selection were performed on nitrous oxide (N₂O) breath spectra obtained from a distributed-feedback quantum-cascade-laser-based multi-pass absorption spectroscopy system. The analysis included 592 spectra from the healthy cohort and 407 spectra from the patient cohort with endoscopically confirmed gastric disease. The baseline fluctuations observed in the N₂O spectra consisted of two main components: a slowly varying trend and an oscillatory pattern. Therefore, a deterministic baseline correction model combining linear and sinusoidal components was applied. Raw and baseline-corrected spectra were compared with a software-derived Voigt reconstruction, revealing that in both cohorts, the root mean square error (RMSE) decreased by approximately 25-28% and the signal-to-noise ratio increased by approximately 19-27%. Local line-shape modeling was then performed using Lorentzian, pseudo-Voigt, and exponentially modified Gaussian profiles under both fixed and signal-to-noise ratio-optimized window conditions. Under fixed-window conditions, the exponentially modified Gaussian model was selected as the best-performing profile based on the lowest local RMSE, in 433 healthy and 281 patient spectra. The area under the curve-to-ppm calibration analysis supported its quantitative consistency, yielding R² = 0.6970 and RMSE = 0.0905 ppm in the healthy group, and R² = 0.8980 and RMSE = 0.0903 ppm in the patient group. These findings indicate that in the present dataset, line-shape selection influenced N₂O spectral fitting and concentration reconstruction more clearly than moderate fitting-window optimization. The proposed framework offers a computationally practical and physically interpretable approach for modeling N₂O breath spectra and supports the use of exponentially modified Gaussian fitting for MuPAS-based breath analysis.