Deep Learning Image Analysis of Nanoplasmonic Sensors: Toward Medical Breath Monitoring

Sensing biomarkers in exhaled breath offers a potentially portable, cost-effective, and noninvasive strategy for disease diagnosis screening and monitoring, while high sensitivity, wide sensing range, and target specificity are critical challenges. We demonstrate a deep learning-assisted plasmonic sensing platform that can detect and quantify gas-phase biomarkers in breath-related backgrounds of varying complexity. The sensing interface consisted of Au/SiO₂ nanopillars covered with a 15 nm metal-organic framework. A small camera was utilized to capture the plasmonic sensing responses as images, which were subjected to deep learning signal processing. The approach has been demonstrated at a classification accuracy of 95 to 98% for the diabetic ketosis marker acetone within a concentration range of 0.5-80 μmol/mol. The reported work provides a thorough exploration of single-sensor capabilities and sets the basis for more advanced utilization of artificial intelligence in sensing applications.