Espectroscopia no infravermelho por transformada de Fourier modo de reflexão total atenuada (ATR-FTIR) como plataforma para detecção de Salmonella Typhimurium

Abstract

Salmonella is a significant pathogen implicated in numerous foodborne diseases. This research integrates attenuated total reflectance-Fourier transform infrared spectroscopy (ATR-FTIR) with artificial intelligence (AI) to identify Salmonella Typhimurium (ST). Infrared (IR) spectra were collected from five Salmonella serotypes: ST, S. Gallinarum (SG), S. Enteritidis (SE), S. Heidelberg (SH), and S. Dublin (SD). The dataset was split randomly into 155 samples for training and 120 samples for external validation to develop a comprehensive Salmonella database. The machine learning models demonstrating superior predictive performance included Random Forest, Support Vector Machine (SVM), and Neural Network. Random Forest achieved a remarkable accuracy of 100.00% during cross-validation and 98.33% in external validation. The study identified Random Forest for external validation (sensitivity: 100.00%, specificity: 98.00%, AUC: 99.5%) and SVM (sensitivity: 100.00%, specificity: 96.00%, AUC: 98.5%) as the most effective models for distinguishing ST from other Salmonella serotypes. The findings underscore that combining ATR-FTIR with machine learning offers a rapid, cost-effective, and on-site applicable diagnostic approach, presenting substantial promise for Salmonella detection.