An Optical Sensing System for Exhaled NH3 Detection Based on Adaptive Envelope Differential Enhancement Spectroscopy Combined with Convolutional Neural Networks.

Exhaled ammonia (NH3) is a key biomarker of chronic kidney disease (CKD), and its accurate detection is of great significance for the noninvasive diagnosis and treatment of CKD. Though ultraviolet differential absorption spectroscopy offers potential for NH3 detection, its practical application faces challenges such as cross-sensitivity, noise interference, and baseline drift. Therefore, we report an NH3 sensing system that integrates adaptive envelope differential enhancement spectroscopy with convolutional neural networks (CNN). An adaptive envelope differencing method based on upper and lower envelope extremum detection is established in the system, which adaptively tracks local variations in slow-varying components to remove their interference and suppress baseline drift, thereby ensuring spectral stability. Subsequently, feature-enhanced spectroscopy is proposed to eliminate narrow-band absorption and noise interference from exhaled breath, which significantly enhances the characteristic signal of NH3. Finally, the CNN is adopted for concentration inversion to achieve quantitative NH3 detection. The results indicate that the sensor system has an R2 of 0.999 and a MAPE of 0.64% in the range of 78.83-7653.13 ppb. The detection limit is 10 ppb at an effective optical path length of 3.2 m. Moreover, despite the presence of multiple interfering components, the system demonstrates good stability and repeatability in detecting exhaled NH3 from simulated CKD patients. This work offers new insights into the application of broadband spectroscopic technology in the noninvasive diagnosis and disease progression monitoring of CKD.