Surface-Engineered Carbon Nanotubes Aerogel Sensor Array Enabled AI-Driven Selective Detection of Biomarker VOCs: A...

Selective detection and quantitative analysis of biomarker volatile organic compounds (VOCs) in complex gas mixtures remain a major challenge for noninvasive breath diagnostics due to wide concentration variability and strong humidity interference. Here, a deliberately heterogeneous chemiresistive sensor array based on six flexible, freestanding porous carbon nanotube (CNT) aerogel films is reported, where the CNT aerogel simultaneously functions as a conductive ambient-temperature transducer and a high-surface-area scaffold. Selectivity was engineered through integration with metal oxides (CuO, V₂O₅, TiO₂, WO₃) and graphitic carbon nitride (g-C₃N₄), generating diverse electronic structures and adsorption kinetics that produce information-rich response signatures. The array was evaluated using six structurally distinct VOCs (acetone, methanol, ethanol, hexane, toluene, and benzene) under varying humidity environments (RH 40-90%). To overcome environmental drift, multidimensional feature vectors were derived from both steady-state response magnitudes and transient kinetic slopes. Correlation analysis revealed weakly correlated and nonlinear relationships among sensor features, motivating the development of a dual-branch deep neural network for simultaneous VOC classification and concentration prediction within a defined, nested multicomponent mixture framework. The model achieved near-perfect classification accuracy and precise quantitative prediction (R² > 0.98) across the discrete composition levels dictated by the mixture recipes. Under simulated breath-biomarker matrices featuring multiple competing VOCs and high background relative humidity (up to 90% RH), the platform demonstrated robust fingerprint deconvolution, achieving ∼99% acetone quantification accuracy within the evaluated mixture templates. While cross-reactivity against physiological interferents such as CO₂ and NH₃ remains to be evaluated for clinical deployment, this AI-assisted, room-temperature CNT aerogel sensor platform establishes a robust proof-of-concept for the pattern-matching analysis of complex VOC environments, advancing reliable sub-ppm trace chemical tracking for future noninvasive biomarker diagnostics.