Synergistic Integration of Machine Learning with Microstructure/Composition-Designed SnO₂ and WO₃ Breath Sensors.

Area of Science:

• Materials Science
• Biomedical Engineering
• Artificial Intelligence

Background:

• Semiconductor metal oxide (SMO) gas sensors are crucial for disease detection.
• Individual SMO sensors often lack the specificity for complex disease biomarker analysis.
• Machine learning (ML) integration is key to enhancing sensor array performance.

Purpose of the Study:

• To develop a high-performance gas sensing strategy for disease prediction using complementary SMO sensors.
• To investigate the efficacy of hybrid sensor arrays and ML algorithms for distinguishing specific gases.
• To assess the potential for monitoring diet and Irritable Bowel Syndrome (IBS) through breath analysis.

Main Methods:

• Utilized a complementary sensor array comprising tin dioxide (SnO₂) and tungsten trioxide (WO₃)-based sensors.
• Employed supervised learning algorithms, including deep neural networks (DNNs) and convolutional neural networks (CNNs).
• Exposed sensors to gas mixtures containing acetone, hydrogen, and ethanol to simulate breath conditions.

Main Results:

• Hybrid integration of SnO₂ and WO₃ sensors significantly improved discrimination ability compared to individual sensor types.
• DNN and CNN models successfully differentiated between acetone and hydrogen, even with ethanol interference.
• The developed system demonstrated potential for predicting diet status and IBS.

Conclusions:

• Hybrid SMO sensor arrays combined with advanced ML algorithms offer a promising approach for accurate breath analysis.
• This strategy enhances gas discrimination for non-invasive disease prediction.
• The findings pave the way for advanced, high-performance breath sensors in healthcare.