Deep learning-based cross-device standardization of surface-enhanced Raman spectroscopy for enhanced bacterial...

Area of Science:

• Spectroscopy
• Machine Learning
• Biotechnology

Background:

• Surface-enhanced Raman spectroscopy (SERS) offers label-free pathogen detection but faces challenges in clinical diagnostics.
• Inconsistent spectral quality, poor reproducibility, and limited machine learning generalizability hinder SERS adoption.
• These limitations impede reliable, rapid pathogen identification at the point of care.

Purpose of the Study:

• To develop a deep learning framework to enhance SERS spectral quality from portable devices.
• To improve the accuracy and generalizability of machine learning models for pathogen classification.
• To enable reliable, real-time pathogen identification at the point of care.

Main Methods:

• Collected SERS spectra from 20 analyte classes using portable and laboratory-grade Raman systems.
• Developed SERS-D2DNet, a sequence-to-sequence network, to transform portable SERS spectra into laboratory-grade equivalents.
• Implemented SuperRaman, a lightweight super-operational neural network, for multiclass bacterial classification.

Main Results:

• SERS-D2DNet significantly improved spectral quality, reducing mean absolute error to 0.01 and increasing R² to over 98%.
• SuperRaman achieved up to 100% classification accuracy after spectral transformation.
• The combined framework demonstrated superior performance over existing methods with a compact footprint and fast inference time.

Conclusions:

• The proposed deep learning framework bridges the performance gap between portable and laboratory-grade SERS systems.
• This scalable, real-time solution facilitates rapid sepsis detection and pathogen identification.
• The technology is well-suited for portable deployment, advancing point-of-care diagnostics.