The paper presents a schematic diagram of the experimental setup human breath analysis using a broadband infrared quantum-cascade laser and a multipass Herriot gas cell. The sensitivity of described experimental setup allows to detect most common biomarkers of various diseases. The experimental spectra for several test substances are measured, threshold concentrations are detected. Statistical numerical methods and deep learning for spectra analysis are considered.
In this work, the possibility of using machine learning in the spectral analysis of exhaled breath for early diagnosis of diseases is considered. Experimental setup consists of a quantum cascade laser with a tuning range of 5.4-12.8 μm and Herriot astigmatic gas cell. A shallow convolutional neutral network and principal component analysis is used to identify biomarkers and its mixtures. A minimum detectable concentration for acetone and ethanol at sub-ppm level is obtained for optical path length up to 6 m and signal-to-noise less than 3. It is shown that neural networks in comparison with statistical methods give a lower detection limits for the same signal-to-noise ratio in the measured spectrum.
We consider the possibility of using a combination of machine and deep learning in the spectral analysis for multicomponent gas mixtures. The experimental setup consists of a quantum cascade laser with a tuning range of 5.3 to 12.8 μm, a peak power up to 150 mW, and a Herriot astigmatic gas cell with an optical path of up to 76 m. We used acetone, ethanol, methanol, acetaldehyde, and ethylene as test substances for the described techniques. For detection and clustering biomarkers, we used machine learning methods such as t-distributed stochastic neighbor embedding, principal component analysis, and classification methods such as decision tree, k-nearest neighbors, logistic regression, and support vector machines. We used a shallow convolutional neural network (CNN) based on TensorFlow (Google) and Keras for spectral analysis of gas mixtures. We modeled IR spectra of pure substances using an NIST database as training and validation sets. Then we used experimental spectra as a test set. We showed that logistic regression gives us the best result for pure substances’ classification. Next, we modeled gas mixtures from synthetic IR spectra for CNN as training and validation sets. We showed that neural networks trained on synthetic spectra can recognize synthetic gas mixtures and experimental individual gaseous substances. We suggest using machine learning methods for pure substances clustering and classification and CNN for gas mixture components identification. We experimentally obtained minimum detectable concentration at ppm level for pure substances. Finally, we estimated that detection limits for the described experimental setup and numerical techniques occurs at levels around 10 to 50 ppb.
In this work, the possibility of using machine learning in the spectral analysis of exhaled breath for early diagnosis of diseases is considered. Experimental setup consists of a quantum cascade laser with a tuning range of 5.4–12.8 μm and Herriot astigmatic gas cell. A shallow convolutional neutral network and principal component analysis is used to identify biomarkers and its mixtures. A minimum detectable concentration for acetone and ethanol at sub-ppm level is obtained for optical path length up to 6 m and signal-to-noise less than 3. It is shown that neural networks in comparison with statistical methods give a lower detection limits for the same signal-to-noise ratio in the measured spectrum.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
INSTITUTIONAL Select your institution to access the SPIE Digital Library.
PERSONAL Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.