A Machine Learning Framework for Detecting COVID-19 Infection Using Surface-Enhanced Raman Scattering.

Area of Science:

• Biomedical Engineering
• Machine Learning
• Spectroscopy

Background:

• Accurate and rapid COVID-19 detection is crucial for public health.
• Surface-Enhanced Raman Scattering (SERS) offers a sensitive method for biological sample analysis.
• Machine learning (ML) can enhance diagnostic capabilities by analyzing complex spectral data.

Purpose of the Study:

• To explore ML approaches for predictive COVID-19 diagnosis using SERS.
• To evaluate dimensionality reduction techniques for SERS data.
• To implement Gaussian process (GP) classification for probabilistic SERS-based diagnosis.

Main Methods:

• Collected SERS data from 20 patients, with labels confirmed by Polymerase Chain Reaction (PCR).
• Compared linear and nonlinear dimensionality reduction methods for high-dimensional Raman spectra.
• Utilized 10-fold cross-validation to determine optimal feature reduction.
• Applied Gaussian process (GP) classification for probabilistic prediction of COVID-19 infection.

Main Results:

• Dimensionality reduction techniques successfully projected high-dimensional SERS data into a lower-dimensional space.
• GP classification accurately predicted sample status (positive/negative) with associated probabilities.
• The framework demonstrated potential for high-throughput, rapid screening.

Conclusions:

• ML-driven SERS analysis provides a viable framework for rapid, probabilistic COVID-19 detection.
• This approach can supplement traditional PCR testing, especially for high-throughput screening.
• The probabilistic output of GP classification allows for informed decisions on confirmatory testing.