Using Machine Learning and Silver Nanoparticle-Based Surface-Enhanced Raman Spectroscopy for Classification of...

Area of Science:

• Biomedical Engineering
• Analytical Chemistry
• Computational Biology

Background:

• Surface-enhanced Raman spectroscopy (SERS) combined with machine learning (ML) offers potential for point-of-care disease detection.
• ML models require adequate training data to avoid generating false or non-generalizable classifications.
• Understanding how SERS experimental parameters affect ML classification is critical for clinical applications.

Purpose of the Study:

• To investigate the impact of SERS methodologies and workflow parameters on ML-based disease classification.
• To assess the accuracy of label-free SERS and ML in classifying simulated cardiovascular disease (CVD) samples.
• To identify factors influencing ML classification performance in SERS analysis of clinical samples.

Main Methods:

• Utilized a label-free, Ag nanoparticle-based SERS platform with ML for analyzing simulated CVD samples in serum.
• Collected Raman spectra at 532, 633, and 785 nm from up to 300 unique samples.
• Employed two standard ML models to classify samples into physiological and pathological categories.

Main Results:

• High classification accuracies (up to 90.0%) were achieved at 532 nm using as few as 200 training samples.
• Single-wavelength SERS at 532 nm yielded the highest accuracies; multiwavelength SERS offered no significant improvement.
• Experimental variables like substrate lots and sample handling strongly influenced ML classification, potentially inflating accuracies by up to 27%.

Conclusions:

• Label-free SERS coupled with ML demonstrates high accuracy for CVD classification.
• Standardized SERS experimental protocols and robust data acquisition are essential for reliable ML-driven diagnostics.
• This study provides insights for improving SERS data sets for practical point-of-care testing applications.