Breath air analysis using wide-band tuning range IR laser photoacoustic spectroscopy and machine learning

Breath air analysis using wide-band tuning range IR laser photoacoustic spectroscopy and machine learning

Yury V. Kistenev 1,2,*, Alexey V. Borisov 1,2, Dmitry A. Kuzmin 2, Denis A.Vrazhnov 3, Olga V. Penkova 1

1. National Research Tomsk State University, 36 Lenin av., Tomsk, Russia;
2. Siberian State Medical University, 2 Moscowski Trakt St., Tomsk, Russia;
3. Institute of Strength Physics and Materials Science SB RAS, Tomsk, Russia

Poster PDF

Abstract: 

The infrared laser photoacoustic spectroscopy (LPAS) abilities and the pattern-recognition-based approach for non-invasive express diagnostics of pulmonary diseases based on absorption spectra analysis of the patient’s breath air are discussed. The study was involved with lung cancer patients (N=30), patients with chronic obstructive pulmonary disease (N= 40), pneumonia (N= 40), and a control group of 130 healthy non-smoking volunteers. The analysis of measured spectra was based, at first, on the reduction of the dimension of the feature space using Principal Component Analysis. Then, the multi-group One-Vs-One classification has been carried out using Support Vector Machine. The method of gas-chromatography-mass-spectrometry (GC-MS) was used as a reference one. The estimated sensitivity of breath air samples analysis with the LPAS in dichotomous classification was not worse than 86%, and the specificity was not worse than 83%. The analogous results in dichotomous classification with GC-MS were 68% and 60%, correspondingly.
Other applications will also be discussed.