A comprehensive approach to evaluating therapy and controlling asthma symptoms in children using the method of Fourier-transform infrared spectroscopy.

Abstract

Childhood bronchial asthma remains a leading chronic pathology in Russia, with up to 40% of patients uncontrolled due to the lack of objective, effort-independent tools for monitoring airway inflammation. This work pioneers a pediatric breath analysis system using mid-infrared Fourier transform spectroscopy with deep cryogenic water suppression (10⁴-fold reduction), enabling simultaneous sub-ppm detection of NO, COS, CO₂, NH₃, and H₂CO. For the first time, a multi-scale convolutional neural network and Spectral Transformer extract breath spectroscopic patterns linked to asthma phenotypes; attention maps physically highlight discriminative absorption lines (e.g., NO ~1900 cm⁻¹, COS ~2050 cm⁻¹) for interpretable AI. A cross-concentration generalization approach maintains >88% classification accuracy across a 10-fold concentration range, critical for age-related variability. The domestic system exceeds global analogues in multigas capacity and noise immunity (>99% accuracy at SNR 5 dB). A verified database of IR breath profiles (120 asthmatic children, 30 healthy controls) will be built, threshold values for NO and COS defined for ages 6–17, and prognostic models developed with ≥92% asthma control accuracy (versus ACT) and early risk identification. Pulmonologist-validated attention maps ensure physical interpretability, establishing pediatric infrared breath spectroscopy for personalized monitoring and early exacerbation prevention, aligned with Russia’s personalized medicine priority.

Speaker

Igor Golyak
Bauman Moscow State Technical University
Russia

Discussion

Ask question