This study explores the potential of non-invasive lung cancer diagnosis using volatile organic compounds (VOCs) in exhaled breath. The researchers analyzed breath samples from lung cancer patients and healthy volunteers using gas chromatography-mass spectrometry (GC-MS), identifying 205 VOCs. Statistical analysis, focusing on VOC ratios rather than individual peak areas, revealed several ratios significantly correlated with lung cancer status, tumor localization, TNM stage, and treatment status. Gradient boosted decision trees (GBDT) and artificial neural networks (ANN) were employed to create diagnostic models, with ANN demonstrating superior accuracy on the test dataset (82-88% sensitivity, 80-86% specificity). The study also highlighted the influence of comorbidities on VOC profiles, emphasizing the need to account for these factors in future diagnostic model development.
