BackgroundIn this study, exhaled breath testing has been considered a promising method for the detection and monitoring of breast cancer (BC). MethodsA high-pressure photon ionization time-of-flight mass spectrometry (HPPI-TOFMS) platform was used to detect volatile organic compounds (VOCs) in breath samples. Then, machine learning (ML) models were constructed on VOCs for the diagnosis of BC and its progression monitoring. Ultimately, 1981 women with useable breath samples were included in the study, of whom 937 (47.3 %) had been diagnosed with BC. VOC panels were used for ML model construction for BC detection and progression monitoring. ResultsOn the blinded testing cohort, this VOC-based model successfully differentiated patients with and without BC with sensitivity, specificity, and area under receiver operator characteristic curve (AUC) values of 85.9 %, 90.4 %, and 0.946. The corresponding AUC values when differentiating between patients with and without lymph node metastasis (LNM) or between patients with tumor-node-metastasis (TNM) stage 0/I/II or III/IV disease were 0.840 and 0.708, respectively. While developed VOC-based models exhibited poor performance when attempting to differentiate between patients based on pathological patterns (Ductal carcinoma in situ (DCIS) vs Invasive BC (IBC)) or molecular subtypes (Luminal vs Human epidermal growth factor receptor 2 (HER2+) vs Triple-negative BC (TNBC)) of BC. ConclusionCollectively, the HPPI-TOFMS-based breathomics approaches may offer value for the detection and progression monitoring of BC. Additional research is necessary to explore the fundamental mechanisms of the identified VOCs.