Real‐Time and Non‐Invasive Detection of Respiratory Viral Infections Using an Intelligent Odor Monitoring System (IOMS)

ABSTRACT Real‐time monitoring of infection‐associated volatile organic compounds (VOCs) offers a non‐invasive pathway for early respiratory infection detection. However, diagnostic precision remains challenged by complex VOC mixtures, low analyte abundance, and significant biological variability. This study introduces an Intelligent Odor Monitoring System (IOMS) guided by infection‐associated biomarker identification. Identification of infection‐associated markers including ethyl lactate and 3,5‐dimethyloctane informed the development of a high‐sensitivity sensor array integrated into an individually ventilated cage (IVC) platform. Operating at 1 Hz, the system generated 3,628,800 longitudinal measurements over a 7‐day infection cycle, capturing distinct odor‐response dynamics between infected and uninfected groups while exploratory principal component analysis supported stage‐associated temporal patterns. Machine‐learning models, including KNN, SVM, and LDA, leveraged these temporal dynamics to model the complete infection cycle with the best internal accuracy reaching 99.88% and model generalizability further supported by an independent blinded cohort. Notably, early sensor‐response shifts were observed at 7–8 hours post‐infection (hpi), and robust model discrimination was achieved by 9 hpi, supporting the feasibility of early‐stage infection monitoring in this murine model. By fusing biomarker‐guided odor sensing with machine learning, the IOMS enables autonomous, continuous surveillance and highlights the potential of longitudinal VOC monitoring for preclinical respiratory infection studies.