Machine learning-assisted limonene sensor based on hierarchical WO3 polyhedrons for non-Invasive diagnosis of liver disease via exhaled breath analysis

Non-invasive early detection of liver disease via breath analysis is a long-sought, yet challenging, target in sensor research. High humidity, complex chemical composition, and substantial inter-individual differences in exhaled breath significantly complicate the breath analysis for liver disease diagnosis. In this work, hierarchical WO 3 polyhedrons were successfully prepared via a simple hydrothermal method. The as-prepared WO 3 material showed the highest ever-reported response (268.2) to 10 ppm limonene (a critical biomarker for liver diseases) at 100°C, with a remarkable detection limit of 10 ppb. Various characterization techniques, including O 2 -TPD, XPS, and gas chromatography-mass spectrometry (GC-MS), were employed to thoroughly analyze the sensing mechanism of WO 3 toward limonene. Based on a comprehensive understanding of the entire sensing process, a robust few samples with high feature dimensionality orient machine learning algorithm was used to assist the sensor in decoding response features specific to limonene from complex breath analysis response singles, thereby achieving 99.9% correct classification rate for distinguishing exhaled signals from liver disease patients and healthy controls in 200 independent cross-validation tests. This work presents the first successful integration of machine learning with a highly sensitive WO 3 -based limonene sensor, offering significant potential for non-invasive, real-time, and cost-effective early detection of liver diseases.