Exhaled Breath Analysis for Detection of Interstitial Lung Diseases Using Real-Time PTR-TOF-MS

Interstitial lung diseases (ILDs) comprise a heterogeneous group of diffuse parenchymal lung disorders requiring accurate non-invasive diagnostic approaches. This study evaluated real-time proton-transfer-reaction time-of-flight mass spectrometry (PTR-TOF-MS) combined with machine learning for breath-based ILD detection and characterization. Overall, 251 patients with ILD and 251 healthy controls underwent analysis of volatile organic compounds (VOCs) during normal quiet breathing and forced expiration. The ILD cohort comprised multiple clinically heterogeneous ILD subtypes. Gradient boosting models were evaluated using repeated Monte Carlo cross-validation, with additional age- and sex-matched validation, ILD subtype classification, and kernel canonical correlation analysis (KCCA). Models incorporating VOCs and demographic variables achieved AUCs of 0.900 and 0.897 for normal and forced breathing, respectively, while VOC-only models retained good discrimination (AUC 0.874 and 0.875). Four VOC-related ions (m/z 55.03859, 63.01996, 90.95284, and 107.0886) were reproducibly identified. Discrimination among individual ILD subtypes was modest, with the highest AUC observed for IPF (0.709). KCCA demonstrated significant associations between VOC signatures and clinical-functional impairment under both breathing conditions. These findings support PTR-TOF-MS breath analysis as a promising non-invasive approach for ILD detection, while subtype classification and disease severity assessment require further independent validation.