Abstract
The opioid epidemic continues to escalate, with a significant number of fatalities occurring when individuals are alone and unable to seek help. This paper presents OxyCaps, a novel shoulder-worn wearable system designed for real-time opioid overdose detection. By integrating reflectance photoplethysmography (PPG) and inertial measurement units (IMU), the system monitors oxygen saturation, respiration, and motion. Using a Capsule Network architecture and a surrogate dataset of sleep apnea patients, OxyCaps achieves 92% accuracy and a 0.98 recall for severe hypoxemia, offering a discreet and effective solution for overdose intervention.
Introduction
With over 120,000 lives lost annually to the opioid epidemic, there is a critical need for automated detection systems. Most fatal overdoses occur in isolation, preventing the timely administration of life-saving agents like naloxone. Developing wearable solutions faces two primary challenges: the lack of high-fidelity physiological data from actual overdose events and the social stigma associated with conspicuous medical devices. OxyCaps addresses these by utilizing a discreet form factor and a "digital twin" data approach.
Methodology
The OxyCaps hardware is a custom sensor adhered to the deltoid, measuring blood oxygen via reflectance PPG and tracking motion/breathing effort through a 6-axis IMU.
The software utilizes Capsule Networks (CapsNets), which employ a dynamic routing mechanism to preserve spatial and temporal relationships between features. This is particularly effective for modeling the lag between respiratory depression and subsequent hypoxemia. The pipeline includes optical normalization for diverse skin tones, multi-modal feature extraction, and a three-tier classification of hypoxemia severity (Normal, Moderate, Severe).
Data Acquisition and Surrogate Modeling
To overcome the "data desert" of illicit overdose events, the researchers used Obstructive Sleep Apnea (OSA) as a physiological proxy. OSA shares key biomarkers with opioid overdoses, including apnea, motionlessness, and oxygen desaturation. The model was trained on over 150 hours of raw data from 19 patients, with ground truth established by FDA-approved medical monitors.
Results and Discussion
OxyCaps demonstrated high performance across all metrics, with an average accuracy of 92%. Crucially, for the "Severe" hypoxemia state—the most vital for overdose detection—the model achieved a recall of 0.98, ensuring minimal missed life-threatening events.
A key finding was the critical role of multi-modal sensing. Incorporating respiration rate (RR) alongside SpO2 significantly boosted performance.
As shown in the table above, adding RR features improved the model's ability to identify hypoxemia by approximately 31% compared to SpO2 alone. This multi-modal approach is essential for reducing false positives and building user trust.
Conclusion and Future Work
OxyCaps represents a significant step toward a discreet, reliable overdose detection system. Future work will focus on expanding the dataset to include more diverse populations and validating the system in controlled clinical settings using drug-induced apnea tests. The ultimate vision is a "closed-loop" system that integrates detection with an automated naloxone auto-injector.
Summary updated