AI-Based Quantification of Volatile Organic Compounds in Gas Mixtures Exploiting Temperature Modulation of MOS Sensor Array

The analysis of biological fluids, particularly exhaled breath (EB), offers a promising noninvasive approach for early disease diagnosis by detecting volatile organic compounds (VOCs). Traditional techniques like gas chromatography-mass spectrometry (GC-MS), while sensitive, are limited by high costs and complexity. This study explores the application of temperature modulation (TM) to enhance the performance of metal oxide semiconductor (MOS) sensors in identifying and quantifying VOCs in complex mixtures. Using a custom-built electronic nose (eNose) system equipped with four MOS gas sensors and a square-triangular TM pattern, mixtures of three VOCs, namely acetone, isopropanol, and toluene, were analyzed across three concentration ranges. Sensible parameters extracted from each sensor response were used to discriminate VOCs and concentrations by random forest (RF) classifier achieving an accuracy of 91%, precision of 91%, recall of 89%, and F1 -score of 89% in classifying the mixtures. Feature re-mapping coupled with a CatBoost classifier leveraging individual VOC analysis reduced the experimental burden and achieved an 84% classification accuracy. These findings demonstrate that TM combined with artificial intelligence (AI) can address key challenges in complex gas mixture analysis, advancing the potential of portable eNose systems for clinical diagnostics.