Prediction of oral squamous cell carcinoma based on machine learning of breath samples: a prospective controlled study

Abstract

Abstract Background The aim of this study was to evaluate the possibility of breath testing as a method of cancer detection in patients with oral squamous cell carcinoma (OSCC). Methods Breath analysis was performed in 35 OSCC patients prior to surgery. In 22 patients, a subsequent breath test was carried out after surgery. Fifty healthy subjects were evaluated in the control group. Breath sampling was standardized regarding location and patient preparation. All analyses were performed using gas chromatography coupled with ion mobility spectrometry and machine learning. Results Differences in imaging as well as in pre- and postoperative findings of OSCC patients and healthy participants were observed. Specific volatile organic compound signatures were found in OSCC patients. Samples from patients and healthy individuals could be correctly assigned using machine learning with an average accuracy of 86–90%. Conclusions Breath analysis to determine OSCC in patients is promising, and the identification of patterns and the implementation of machine learning require further assessment and optimization. Larger prospective studies are required to use the full potential of machine learning to identify disease signatures in breath volatiles.

Keywords

• Breath analysis
• Head and neck cancer
• Oral squamous cell carcinoma
• Oral cancer
• Machine learning
• Gas chromatography–ion mass spectrometry