A dynamic threshold adaptive gradient weighted baseline correction algorithm of Raman spectra for breast cancer diagnosis

AbstractBreast cancer is the most prevalent malignancy among women worldwide, highlighting the urgent need for real-time and accurate intraoperative diagnostic techniques. Fiber-optic micro-Raman spectroscopy enables non-invasive, label-free, and real-time detection of biomolecular features in tissues, offering promising potential in clinical breast cancer applications. However, raw Raman spectra are often affected by strong autofluorescence baselines, which obscure critical spectral features and compromise subsequent modeling accuracy. To address these challenges, a dynamic threshold adaptive gradient weighted (DT-AGW) baseline correction algorithm is proposed. This method adaptively adjusts penalization weights across wavenumbers based on spectral gradient features, eliminating the need for prior information and enhancing generalizability. A dynamic threshold mechanism improves peak recognition tolerance, increasing robustness to low signal-to-noise ratio spectra. Furthermore, the weighting function incorporates both spectral gradient and fitting residuals to improve iterative correction accuracy. Experimental results demonstrate that DT-AGW achieves lower Root Mean Square Error and higher classification accuracy compared to conventional baseline correction methods. The proposed algorithm shows strong potential for integration into portable intraoperative diagnostic devices and offers improved performance for biomedical spectral data analysis.