FTIR-based profiling of sinusitis: Stabilized feature selection through Boruta and Monte Carlo methods.

Area of Science:

• Biomedical Spectroscopy
• Analytical Chemistry
• Medical Diagnostics

Background:

• Sinusitis diagnosis currently lacks rapid, non-invasive methods.
• Biochemical changes in sinusitis patients may be detectable in serum.
• Fourier-transform infrared (FTIR) spectroscopy offers potential for biochemical analysis.

Purpose of the Study:

• To investigate the utility of serum FTIR spectroscopy for sinusitis diagnosis.
• To identify spectral features indicative of sinusitis.
• To develop a non-invasive diagnostic approach for sinusitis.

Main Methods:

• Serum samples from sinusitis patients and healthy controls were analyzed using FTIR spectroscopy.
• A two-stage feature selection framework (Boruta, Monte Carlo) was employed.
• Support Vector Machine (SVM) classification and dimensionality reduction (PCA, UMAP, t-SNE) were utilized.

Main Results:

• A spectral band between 3400-3550 cm⁻¹ was consistently identified as important for discrimination.
• Eighteen consensus features enabled an SVM to achieve 94% accuracy on a held-out set.
• Distinct clustering of patient and control groups was observed using dimensionality reduction techniques.

Conclusions:

• Serum-based FTIR spectroscopy, coupled with advanced feature selection, shows potential as a non-invasive diagnostic aid for sinusitis.
• The high-wavenumber spectral region (3400-3550 cm⁻¹) robustly carries disease-related biochemical information.
• Further comparative studies are warranted to validate these findings in broader populations.