Transparent, Cost-Sensitive Threshold Calibration for Autonomous Edge-AI Health Wearables: A Severity-Stratified Hysteretic Algorithm for Predictive Asthma Facemasks

Organizations deploying autonomous, AI-enabled health wearables must govern the opaque decision rules that gate their interventions, which carry direct consequences for cost, safety, and accountability. Wearable predictive devices for asthma combine environmental and physiological sensing with on-device actuation, yet these decision thresholds are almost always fixed heuristics. Fixed thresholds ignore inter-patient severity, trade sensitivity against false alarms suboptimally, and cause actuator chattering that wastes battery and metered medication. This paper presents AeroGate, a single-threshold calibration algorithm for a smart, AI-enabled asthma facemask that warms inhaled air, classifies cough acoustics, and triggers a vibrating-mesh nebulizer. AeroGate derives a population severity prior from chest imaging and converts a calibrated risk score into per-patient actuation thresholds through imaging-driven severity stratification, a cost-sensitive base threshold that weighs a missed attack against a false alarm, a data-sized hysteresis band that suppresses chattering, and an O(۱) on-device update that holds a target alarm budget. Evaluated on the NIH ChestX-ray۱۴ benchmark (۱۱۲,۱۲۰ studies) with strict patient-wise splits, the calibrated head attains an AUC of ۰.۶۸ and a Brier score of ۰.۲۳; AeroGate reduces actuator toggling by ۷۷% relative to a single-threshold policy while matching detection sensitivity and holds the realised alarm rate at its target budget. Beyond the engineering, AeroGate reframes the operating point as a transparent, auditable management lever: the alarm budget becomes a cost-and-service target that managers can hold, and the cost ratio becomes a risk-tolerance input that clinicians and managers can set, yielding a principled, low-infrastructure recipe for deploying, governing, and scaling autonomous edge-AI health interventions.

کلیدواژه ها:

Digital health technology management ، Edge-AI governance ، Threshold calibration ، Cost-sensitive decision support ، Predictive wearable devices ، Technology adoption and deployment