Automated Baseline Correction Evaluation Score for Raman Spectroscopy.

Raman spectroscopy is a noninvasive technique providing detailed molecular insights into tissue composition, enabling applications in diagnostics and therapy monitoring. However, Raman signals are inherently weak and prone to superimposed background signals, making baseline correction, a critical step to remove background signals, essential. Typically, experts assess the quality of baseline fitting visually, which is subjective, time-consuming, and impractical with large clinical data sets. To address this, we propose the IS (Integrity Spectrum) Score, a quantitative and automatic metric that evaluates baseline correction quality directly from Raman spectra. The algorithm leverages several spectral properties: it measures the distance between the baseline and the band's prominence, quantifies the Raman shift distances within band edges using a band prominence-proportional factor, compares area-under-the-curve differences between the raw spectrum and an intentionally overfitted baseline, and counts dips where the baseline rises above the spectral curve. These combined indicators capture both underfitting and overfitting behaviors without requiring ground truth labels or manual inspection. Results obtained using complex two-layered blood cell data sets show a strong correlation between IS-Score and expert evaluations, while substantially reducing the time required from experts. This method supports multicenter studies by facilitating consistent baseline correction, reducing labor, and minimizing subjective bias, ultimately enhancing the reliability and comparability of Raman data for clinical research.