Rapid identification of pathogenic bacteria using Raman spectroscopy and deep learning

Area of Science:

• Biomedical Engineering
• Spectroscopy
• Machine Learning

Background:

• Raman optical spectroscopy offers label-free bacterial detection, identification, and antibiotic susceptibility testing.
• Clinical application is limited by weak Raman signals and bacterial diversity, hindering speed and accuracy.

Purpose of the Study:

• To develop a deep learning approach for accurate bacterial identification and antibiotic susceptibility testing using Raman spectra.
• To overcome challenges of weak signals and diverse bacterial phenotypes for clinical relevance.

Main Methods:

• Generated an extensive dataset of bacterial Raman spectra.
• Applied deep learning algorithms to analyze low signal-to-noise spectra for pathogen identification.
• Validated the approach on clinical isolates from patients.

Main Results:

• Achieved average isolate-level accuracies exceeding 82% for identifying 30 common bacterial pathogens.
• Reached 97.0±0.3% accuracy for antibiotic treatment identification.
• Distinguished between methicillin-resistant and -susceptible Staphylococcus aureus (MRSA/MSSA) with 89±0.1% accuracy.
• Demonstrated 99.7% treatment identification accuracy on clinical isolates using minimal spectral data.

Conclusions:

• Deep learning applied to Raman spectroscopy provides a powerful tool for rapid, accurate bacterial diagnostics.
• The method shows significant potential for culture-free pathogen identification and antibiotic susceptibility testing.
• The approach is adaptable for direct analysis of clinical samples like blood, urine, and sputum.