A Review of Machine Learning-Assisted Gas Sensor Arrays in Medical Diagnosis.

Abstract

Volatile organic compounds (VOCs) present in human exhaled breath have emerged as promising biomarkers for non-invasive disease diagnosis. However, traditional VOC detection technology that relies on large instruments is not widely used due to high costs and cumbersome testing processes. Machine learning-assisted gas sensor arrays offer a compelling alternative by enabling the accurate identification of complex VOC mixtures through collaborative multi-sensor detection and advanced algorithmic analysis. This work systematically reviews the advanced applications of machine learning-assisted gas sensor arrays in medical diagnosis. The types and principles of sensors commonly employed for disease diagnosis are summarized, such as electrochemical, optical, and semiconductor sensors. Machine learning methods that can be used to improve the recognition ability of sensor arrays are systematically listed, including support vector machines (SVM), random forests (RF), artificial neural networks (ANN), and principal component analysis (PCA). In addition, the research progress of sensor arrays combined with specific algorithms in the diagnosis of respiratory, metabolism and nutrition, hepatobiliary, gastrointestinal, and nervous system diseases is also discussed. Finally, we highlight current challenges associated with machine learning-assisted gas sensors and propose feasible directions for future improvement.