Clinically Accurate Diagnosis of Asthma via Exhaled Breath Analysis Using an Ultrasensitive NO/H2S Gas Sensor Array Assisted by Machine Learning Analysis

Abstract

Asthma poses a substantial global health burden, affecting approximately 300 million individuals worldwide. However, accurate diagnosis of asthma remains a significant clinical challenge, primarily due to the heterogeneous and dynamic nature of the disease and the lack of a single definitive biomarker. These limitations motivate the continued development of advanced diagnostic strategies, particularly those based on exhaled breath analysis. Since both NO and H2S have been identified as breath biomarkers of asthma, the simultaneous analysis of NO and H2S in the exhaled breath using a gas sensor array composed of ultrasensitive single sensors may enable more accurate clinical diagnosis. In this study, a gas sensor array consisting of an ultrasensitive NO sensor and an ultrasensitive H2S sensor, each with a detection limit of 5 ppb, was constructed. Clinically, 76 exhaled breath samples collected from 36 asthma patients and 40 healthy controls were analyzed using the proposed gas sensor array. Two diagnostic models for asthma were constructed using the machine learning analysis algorithm: one based on the selection and analysis of hand-crafted (expert-defined) features and the other based on analyzing the integrated hand-crafted features with automatically learned features. The two models achieved accuracies, sensitivities, and specificities of 93, 92, and 95% and 91, 89, and 93%, respectively, with corresponding area under the curve values of 0.98 and 0.97. These results demonstrate the strong capability of the proposed gas sensor array for accurate asthma diagnosis and highlight its significant potential for practical implementation in clinical settings.