A Wearable Sensor-Driven Monitoring Framework for Hypoxic Respiratory Key Parameters Using Sensitivity-Based Estimation

Abstract

Wearable sensors provide continuous physiological measurements, such as $\text{SpO}_{2}$ and heart rate. However, in hypoxic respiratory applications, such sensors primarily monitor superficial symptoms, while the key underlying parameters remain nonmeasurable. To address this issue, this work proposes a process monitoring framework driven by a novel sensitivity-based parameter estimation approach. The framework utilizes a physics-informed model with 18 key parameters as the monitoring target, where the change in two critical parameters is used to assess the hypoxic adaptation index (HAI). The estimation method incorporates a sensitivity-based groupwise optimization and an identifiability-based collective variational inference, which assigns greater trust weights to high-quality signals to mitigate the impact of low-quality data. Monitoring effectiveness is validated via an in silico experiment, which demonstrates normalized mean absolute error reductions of 3.28% (individual) and 4.30% (population) over the baseline. The robustness of the framework and the effectiveness of the HAI are subsequently confirmed in a real-world intermittent hypoxic training experiment that involved noisy sensor data from 19 human subjects. This work provides a robust and generalizable framework for monitoring nonmeasurable key parameters in other complex dynamic systems.