Detecting CH4 in breath and its related diseases
Detecting CH4 in breath and its related diseases
Tahereh Shah & Prof Paul Maguire
Full Poster:
Abstract:
Methods
Firstly, before proceeding with breath analysis, we have investigated supervised learning using Partial Least Squares Discriminant Analysis (PLS-DA) on a gas mixture dataset of He/CH4 spectra where the CH4 concentration varies from 0 – 100 ppm. He/CH4 is a complex gas mixture but is still simpler than external clinical or environmental samples. The data was collected in a matrix of 3648 variables (wavelengths), which form 9 CH4 concentration categories (0, 1, 2, 4, 6, 12, 23, 77, 100 ppm). Later, we have tried all achieved result into another experiment. In this method, exhaled breath was collected from five participants. Spectra were collected using an Ocean Optics HR4000CG-UV-NIR spectrometer in the wavelength range 194 – 1122 nm (interval 0.25 nm), with a slit width of 5 mm and a minimum optical resolution >0.5 nm.
Results
From a computation perspective, the major difficulty of CH4/He spectra data is the high feature dimensionality, along with temporal instability or drift, collinearity and a high matrix component. These challenges decreased after pre-processing of spectra to include autoscaling, smoothing and baseline correction, followed by data segmentation, VIPs selection and peak concatenation. As a result, spectral features corresponding to helium, carbon, hydrogen and impurities (N, O, OH/H2O) were observed and the algorithm accuracy on this data improved to 98% with < 15 LV.
Comments
Please sign in to leave a comment.