Estimating baselines of Raman spectra based on transformer and manually annotated data.

Area of Science:

• Analytical Chemistry
• Spectroscopy
• Biophysics

Background:

• Raman spectroscopy is vital for analyzing biological tissues.
• Baseline and noise interferences hinder accurate spectral analysis.
• Current baseline correction methods are often manual and time-consuming.

Purpose of the Study:

• To develop an automated and accurate baseline correction method for Raman spectra.
• To evaluate a deep learning approach against traditional methods.
• To enhance the pre-processing of Raman spectral data for biological applications.

Main Methods:

• Manually curated ground-truth baselines for eight biological materials.
• Tuning parameters of Modified Multi-Polynomial Fit (Modpoly), Improved Modified Multi-Polynomial Fitting (IModpoly), and airPLS methods.
• Designing and implementing a one-dimensional Transformer (1dTrans) model for baseline estimation.
• Comparative analysis against Convolutional Neural Network (CNN) and ResUNet models.

Main Results:

• The 1dTrans model demonstrated superior performance in baseline correction.
• Lower Mean Absolute Error (MAE) and Spectral Angle Mapper (SAM) scores were achieved by 1dTrans.
• The deep learning model proved effective across diverse biological materials.

Conclusions:

• The 1dTrans model offers a versatile and effective solution for Raman spectral baseline correction.
• Automated deep learning methods can overcome limitations of traditional parametric approaches.
• This advancement facilitates more reliable quantitative analysis of biological samples using Raman spectroscopy.