Abstract
The absence of specific markers makes early detection of bladder cancer (BC) challenging. Recent studies have reported the reliable diagnostic significance of volatile organic compounds (VOCs) in several cancers. This study aimed to investigate whether urinary VOCs could serve as potential biomarkers for BC. A total of 89 BC patients and 67 healthy individuals were recruited for this study. VOCs in urine samples were detected using gas chromatography-ion mobility spectrometry (GC-IMS). Machine learning algorithms were used to establish diagnostic models for predicting BC based on differential expressed VOCs. Compared with healthy individuals, A total of 20 differentially expressed VOCs were identified, including 17 upregulated and 3 downregulated. Among five machine-learning algorithms, the Random Forests (RF) provided the highest accuracy. Based on RF analysis, eight important VOCs (2-Undecenal, 2-propanone, 1-octanal, Ethyl 2-hydroxybenzoate, 2-ethyl hexanol-D, 2-Pentanone, 1-Propanol,3-(methylthio), 1-nonanal) were identified and used to constructed a diagnostic model, showing the highest diagnostic accuracy and area under the curve (AUC) of 91.5% and 0.958, respectively. Our results suggest a VOCs panel might be used as novel biomarker for early detection of BC.