Machine Learning-Enabled Gas Sensor Based on MOF-Derived In₂O₃-CuO for Exhaled CO Detection

Area of Science:

• Materials Science
• Chemical Sensing
• Biomedical Engineering

Background:

• Neonatal jaundice screening requires precise, noninvasive methods for detecting carbon monoxide (CO) in breath.
• Existing methods face challenges in specifically identifying CO within complex breath matrices.

Purpose of the Study:

• To develop an electronic nose system for accurate, noninvasive detection of CO as a biomarker for neonatal jaundice.
• To overcome cross-sensitivity issues with interfering gases in clinical breath analysis.

Main Methods:

• Fabrication of In2O3-CuO sensors from bimetallic metal-organic frameworks (Bi MOFs).
• Assembly of a microsensor array and construction of an electronic nose system with linear discriminant analysis.
• Application of thermodynamic feature engineering and machine learning for data analysis and algorithm optimization.

Main Results:

• The five-sensor In2O3-CuO electronic nose accurately discriminated between CO and acetone.
• A single sensor demonstrated high reproducibility, moisture resistance, stability, a 1 ppm detection limit, and rapid response/recovery times (5/21 s).
• The system achieved 82% cross-validation accuracy for qualitative CO identification (1-50 ppm), addressing cross-sensitivity with C3H6O.

Conclusions:

• The developed electronic nose provides a reliable technical platform for noninvasive screening of neonatal jaundice.
• Thermodynamic feature-driven machine learning enhances sensor array performance, reducing data and computational demands.
• This approach offers a promising solution for early detection of bilirubin metabolic abnormalities.