Adaptive Physics-Aware Raman Baseline Correction with Machine Learning Predicted Parameters.

Area of Science:

• Spectroscopy
• Chemometrics
• Machine Learning

Background:

• Accurate baseline correction is essential for interpreting Raman spectra.
• Existing methods have limitations in automation, generalizability, and user control.

Purpose of the Study:

• To develop an automated, adaptive Raman spectral baseline correction method.
• To improve the robustness and accuracy of baseline correction for high-throughput applications.

Main Methods:

• Developed DIRAS (Dynamic Iterative Reweighted Autoregressive Spectral baseline correction) with a fixed regularization parameter (λ).
• Utilized Structural Similarity Index Measure (SSIM) for λ optimization.
• Trained a deep learning model (DIRAS+) for real-time, spectrum-specific λ prediction.

Main Results:

• DIRAS+ demonstrated superior performance over ALS and SEALS on SERS datasets.
• Achieved better peak fidelity, reduced intraclass variability, and minimized baseline distortion.
• DIRAS improved calibration and chemometric model performance, enhancing analytical sensitivity.

Conclusions:

• DIRAS and DIRAS+ provide robust, scalable, and user-adaptable solutions for Raman spectroscopy.
• The methods automate baseline correction, crucial for high-throughput spectral analysis.
• DIRAS+ enables real-time, spectrum-specific baseline correction, advancing spectral interpretation.