An estimated 10.5% of the world’s population aged 20–79 years are currently living with diabetes in 2021. An urgent task is to develop a non-invasive express-diagnostics of diabetes with high accuracy. Type 1 diabetes mellitus (T1DM) diagnostic method based on infrared laser spectroscopy of human exhaled breath is described. A quantum cascade laser emitting in a pulsed mode with a peak power of up to 150 mW in the spectral range of 5.3–12.8 μm and Herriot multipass gas cell with an optical path length of 76 m were used. We propose a method for collecting and drying an exhaled human air sample and have measured 1200 infrared exhaled breath spectra from 60 healthy volunteers (the control group) and 60 volunteers with confirmed T1DM (the target group). A 1-D convolutional neural network for the classification of healthy and T1DM volunteers with an accuracy of 99.7%, recall 99.6% and AUC score 99.9% was used. The demonstrated results require clarification on a larger dataset and series of clinical studies and, further, the method can be implemented in routine medical practice.

<!-- *** Custom HTML *** --> 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.

Greenhouse gases absorb the Earth’s thermal radiation and partially return it to the Earth’s surface. When accumulated in the atmosphere, greenhouse gases lead to an increase in the average global air temperature and, as a result, climate change. In this paper, an approach to measuring CO${}_2$ and CH${}_4$ concentrations using Fourier transform infrared spectroscopy (FTIR) is proposed. An FTIR spectrometer mockup, operating in the wavelength range from 1.0 to 1.7 $\mathsf{\mu }$m with a spectral resolution of 10 cm${}^{-1}$, is described. The results of CO${}_2$ and CH${}_4$ observations throughout a day in urban conditions are presented. A low-resolution FTIR spectrometer for the 16U CubeSat spacecraft is described. The FTIR spectrometer has a 2.0–2.4 $\mathsf{\mu }$m spectral range for CO${}_2$ and CH${}_4$ bands, a 0.75–0.80 $\mathsf{\mu }$m range for reference O${}_2$ bands, an input field of view of 10${}^{-2}$ rad and a spectral resolution of 2 cm${}^{-1}$. The capabilities of the 16U CubeSat spacecraft for remote sensing of greenhouse gas emissions using a developed FTIR spectrometer are discussed. The design of a 16U CubeSat spacecraft equipped with a compact, low-resolution FTIR spectrometer is presented.

<!-- *** Custom HTML *** --> A real-time automated system for remote substance identification on various surfaces without preliminary sample preparation is presented. In practice, it can be used, for example, as an alerting system to signal the presence of some contaminants. The main components of the system are diffuse reflectance spectra acquisition module, data processing module, and identification module. Development of each module was based on the choice of appropriate devices and algorithms, either existing or newly designed. The experimental setup consists of a quantum cascade laser emitting in the spectral range of 5.3 to 12.8 μm with a HgCdTe photodetector. To achieve better selectivity of substance recognition, identification algorithms were based on the absorption and transmission spectra calculated from the recorded diffuse reflectance spectra. Spectra conversion algorithms employed Kramers–Kronig relations, phase spectra extrapolation, and phase correction. The system was supplied with the recognition database composed of certain commercially available substances. The experiments showed that the usage of transmittance spectra significantly improved the sensitivity of the identification method; the remote identification limit of 30 μg acetylsalicylic acid has been experimentally confirmed. For similar substances, such limit was estimated as 10 μg / cm² at a distance of 1 m.

The recovery of Raman spectra in the 800-1050-nm spectral range from a prototype static Fourier-transform spectrometer is described. Distortion effects introduced by the optical system of the static Fourier-transform spectrometer are studied. A method of correcting optical distortion is suggested. The proposed algorithm is tested using a monochromatic radiation source. A two-dimensional map of period distribution in the distorted interference pattern is plotted. A mapping that enables optical distortions of the two-dimensional interferogram to be corrected is proposed. The correction results are demonstrated by plotting interference fringe profiles. A comparison of various well-known phase correction methods in order to identify an optimal algorithm is presented. A matching criterion for the obtained spectra is proposed. Raman spectrum of a stilbene crystal obtained by a diffraction spectrometer with a resolution of δν = 14 cm-1 is used as a reference.

In this paper, the application of machine learning and deep learning in the spectral analysis of multicomponent gas mixtures is considered. The experimental setup consists of a quantum cascade laser with a tuning range of 5.3–12.8 µm, a peak power of up to 150 mW, and an astigmatic Herriott gas cell with an optical path length of up to 76 m. Acetone, ethanol, methanol, and their mixtures are used as test substances. For the detection and clustering of substances, including molecular biomarkers, methods of machine learning, such as stochastic embedding of neighbors with a t-distribution, principal component analysis and classification methods, such as random forest, gradient boosting, and logistic regression, are proposed. A shallow convolutional neural network based on TensorFlow (Google) and Keras is used for the spectral analysis of gas mixtures. Model spectra of substances are used as a training sample, and model and experimental spectra are used as a test sample. It is shown that neural networks trained on model spectra (NIST database) can recognize substances in experimental gas mixtures. We propose using machine learning methods for clustering and classification of pure substances and gas mixtures and neural networks for the identification of gas mixture components. Using the experimental setup described, the experimentally obtained concentration limits are 80 ppb for acetone and 100–120 ppb for ethanol and methanol. The possibility of using the proposed methods for analyzing spectra of human exhaled air is shown, which is significant for biomedical applications.