Multimodal wearable sensing using optical spectroscopy: real-time opioid overdose detection and blood glucose estimation

Anush Niranjan Lingamoorthy

doi:10.17918/00010992

Wearable health monitoring devices have gained increasing attention due to their potential to provide continuous, real-time data for managing chronic and acute medical conditions. However, existing systems often lack usability, reliability, or accessibility, especially among high-risk populations such as people who use opioids (PWUOs) or individuals living with diabetes. This dissertation addresses these challenges by developing and evaluating two distinct wearable systems that leverage optical spectroscopy for non-invasive monitoring, targeting opioid overdose detection and capillary blood glucose estimation, respectively. While both systems utilize optical principles, they are designed as separate research pillars with unique sensor configurations, algorithmic frameworks, and deployment contexts. The first pillar centers on opioid overdose detection through the DOVE system (Detection of Opioid-related Ventilatory Emergency), a shoulder-mounted, autonomous device designed for PWUOs. Opioid overdose remains a critical public health crisis, with over 120,000 overdose deaths per year, and a significant portion of fatalities occurring when individuals use opioids alone, lacking timely intervention. DOVE addresses this need by integrating multimodal sensors, including photoplethysmography (PPG) for oxygen saturation (SpO₂) measurement, accelerometry for motion detection, and respiration rate estimation. The system's architecture optimizes the challenges of shoulder-based sensing, including low perfusion and high motion artifacts. An adaptive LED intensity normalization technique accommodates diverse skin tones, while a ternary spiking neural network (SNN) enables low-power, hypoxia classification. In feasibility studies, DOVE demonstrated reliable SpO₂ estimation with a mean absolute error (MAE) of 4.91% during controlled hypoxia events, maintaining performance under varying environmental conditions. The system's utility was further evaluated in a real-world field study involving 12 PWUOs over five days, where an attention-based BiLSTM model accurately distinguished opioid use from daily activity with over 91% accuracy. These results validate the feasibility of non-invasive, autonomous overdose detection in real-world settings. The second research pillar focuses on non-invasive glucose monitoring through the GlucoLux system. Diabetes management often requires continuous glucose monitoring (CGM), yet current systems are invasive, expensive, and inaccessible to many, particularly in low-resource settings. GlucoLux leverages optical spectroscopy, utilizing a multi-wavelength LED-photodiode configuration to detect glucose fluctuations non-invasively from the fingertip. The sensor array spans visible and near-infrared (NIR) wavelengths, capturing subtle changes in light absorption corresponding to capillary glucose levels. The system's ensemble learning model, trained on data from a seven-participant study, achieved a mean absolute relative difference (MARD) of 5.4%, with 99.97% of estimates falling within Clarke Grid zones A and B. Cross-validation against commercial CGMs demonstrated consistent accuracy, even across diverse skin tones, highlighting the potential for non-invasive glucose tracking. This dissertation contributes to the field of wearable health monitoring by advancing optical biosensing methods specifically tailored for high-impact, underserved healthcare needs. By co-designing hardware and algorithms with end-user involvement and clinical validation, we bridge a crucial gap between laboratory innovation and real-world application.

Multimodal wearable sensing using optical spectroscopy: real-time opioid overdose detection and blood glucose estimation

Files and links (1)

Abstract

Metrics

Details

Multimodal wearable sensing using optical spectroscopy: real-time opioid overdose detection and blood glucose estimation

Files and links (1)

Abstract

Metrics

Details

Drexel University Social media