TB-SERS analyzer: Analysis tool for tuberculosis prediction based on Raman spectroscopy with machine learning and...

Area of Science:

• Biomedical Engineering
• Computational Biology
• Spectroscopy

Background:

• Raman spectroscopy (RS) and surface-enhanced Raman spectroscopy (SERS) show promise for clinical diagnostics, particularly in tuberculosis (TB) detection.
• Existing diagnostic methods for TB can be time-consuming and invasive.
• There is a need for specialized software to analyze RS/SERS data for TB diagnosis.

Purpose of the Study:

• To develop a user-friendly software tool, TB-SERS Analyzer, for tuberculosis prediction using SERS data.
• To integrate machine learning (ML) and one-dimensional convolutional neural network (1D-CNN) models for automated TB diagnosis.
• To create an accessible platform for rapid, non-invasive TB screening.

Main Methods:

• Development of TB-SERS Analyzer, a Python-based software with a graphical user interface (GUI).
• Establishment of a reference database of 1,000 plasma samples (500 IGRA-positive, 500 IGRA-negative) using interferon-gamma release assay (IGRA).
• Training and optimization of ML and 1D-CNN models using five-fold stratified cross-validation, evaluating seven algorithms for TB classification.

Main Results:

• The 1D-CNN model achieved 82.00% sensitivity and 76.00% specificity in the validation set (n=200).
• In a blinded external test (n=20), the model demonstrated 80.00% sensitivity and 100% specificity.
• TB-SERS Analyzer provides diagnostic reports in under 10 seconds per sample, integrating data extraction, preparation, analysis, and report generation modules.

Conclusions:

• TB-SERS Analyzer is an effective and accessible tool for TB screening, combining SERS technology with ML and 1D-CNN models.
• The software offers high efficiency and rapid results for TB diagnosis.
• TB-SERS Analyzer is freely available on GitHub, facilitating wider adoption and research.