Exhaled human breath analysis using machine learning

Human exhaled breath is predominantly turbulent. This turbulence can help build algorithms to investigate the uniqueness of the extrathoracic airway. Air, during exhalation, is forced to flow out of the human lung through the trachea and oral cavity. We attempt to validate the hypothesis that turbulent exhaled flow carries a signature of the source of generation, i.e., the geometry of the upstream flow region. 1D velocity time series samples measured using hot-wire anemometer from 94 human volunteers make the dataset. Features from the available time-series data were extracted using MFDFA, a technique widely used for determining the fractal scaling properties and long-range correlations in the time series. The features are attributes of the multifractal spectrum of exhaled velocity data. Tuned random forest models were employed in building subject confirmation algorithms. A subject confirmation algorithm tries to verify whether a human subject is the person who they claim to be. The machine learning-based algorithm was found to achieve a good true confirmation rate. The accuracy in the binary classification of a majority of subject combinations was above 68%, which signifies the existence of some uniqueness in an individual's exhaled breath. Such an algorithm will help us explore the area of personalized medicines. This could be used as a tool to characterize or diagnose the variation in extrathoracic morphology over time. For example, it is possible that the turbulence information can be correlated to occlusion in the extrathoracic passage, which is a major source of deposition of aerosolized therapeutics.