Non-Invasive Photonic System for Disease Screening: Fingerprinting Exhaled-Breath Metabolites

Compact, label-free optical transducers are attractive for real-time exhaled-breath screening. We report the characterization and classification performance of a low-power photonic elastic-backscatter sensing system (the Dr. T system) that interrogates injected gas mixtures with a 635 nm, <1 mW coherent diode-laser beam and records the back-scattered light on a two-dimensional silicon position-sensing detector (PSD) as a time series of intensity I and centroid coordinates (x, y) at 10 Hz. Seven calibration data sets were acquired in a sealed 0.11 L chamber—five disease-linked metabolite mixtures in a normal-air balance (CKD, SIBO, cystic fibrosis, COPD, heart disease), a healthy base-gas control, and an ambient control (19 records). Regression showed the beam-deflection axis (y vs x) to be a linear, reproducible common-mode signature of injection flow (slopes 0.18–0.24; R2 up to 0.98), whereas the optical response versus concentration (|I| vs C) is species-specific across a >20-fold slope range, constituting an optical fingerprint. A physically motivated feature space with Mahalanobis classification separated the four diseases carrying substantive metabolites at 100% leave-one-run-out recall; overall seven-class accuracy was 78.9%. Decisively, every substantive-metabolite disease was well separated from a base-gas control undergoing an identical chamber-pressure rise (D = 4.8–13.5), demonstrating that discrimination is driven by molecular optical signatures rather than common-mode pressure. The sensor can detect, discriminate, quantify and classify metabolite mixtures in vitro; signal-to-contrast enhancement for low-cross-section analytes, humidity compensation and radiometric calibration are identified as translation priorities.