Qualitative and Quantitative Analysis of Volatile Molecular Biomarkers in Breath Using THz-IR Spectroscopy and Machine Learning

Exhaled air contains volatile molecular compounds of endogenous origin, being products of current metabolic pathways. It can be used for medical express diagnostics through control of these compounds in the patient’s breath using molecular absorption spectroscopy. The fundamental problem in this field is that the composition of exhaled air or other gas mixtures of natural origin is unknown, and content analysis of such spectra by conventional iterative methods is unpredictable. Machine learning methods enable the establishment of latent dependencies in spectral data and the conducting of their qualitative and quantitative analysis. This review is devoted to the most effective machine learning methods of exhaled air sample absorption spectra qualitative and content analysis. The focus is on interpretable machine learning methods, which are important for reliable medical diagnosis. Also, the steps additional to the standard machine learning pipeline and important for medical decision support are discussed.