Quantitative Detection of Trace Characteristic VOCs of Liver Metabolic Diseases Based on the MEMS Sensor Array

There are a variety of volatile organic compounds (VOCs) gases in human exhalation, and among them isoprene, ethanol, and formaldehyde can be used as biomarkers for liver metabolic diseases. In order to accurately detect these trace‐concentration VOC gases, a sensor array was built with 4 MEMS gas sensors, and one of them was the self‐developed sensor, which has a very high response to isoprene. To improve prediction accuracy of gas concentration, we investigated the convolutional neural network with a Multi‐Expert Temporal Fusion Network (METF‐Net) model based on multitask learning. Based on the MEMS sensor array, the isoprene, ethanol, and formaldehyde at sub ppm level can be correctly identified; their RMSEs of isoprene, ethanol, and formaldehyde are 33.48, 64.01, and 18.84 ppb, and the predicted concentrations with error rates of isoprene, ethanol, and formaldehyde are 6.70%, 6.40%, and 9.42%, respectively. This method has the potential of being applied in the screening of liver metabolic diseases at an early stage.