🧠 Research

AI & Wearable Sensing for Physiological Monitoring

Deep learningVO2/VCO2 estimationTransfer learningTextile-based wearable systemsHeart failure monitoring

Overview

I build deep learning models that turn signals from ordinary wearable sensors into estimates of physiological states that normally require lab equipment. This started with running-related oxygen uptake and carbon dioxide production (VO₂/VCO₂), and is now extending toward cardiac and heart-failure-relevant signals from textile-integrated sensors at the University Health Network (UHN), affiliated with the University of Toronto.

Figures

Wearable sensor placement for data collection

Model architecture / estimation pipeline

Estimated vs. measured VO2/VCO2 validation results

Methodology

Transfer learning (PyTorch) to map wearable-sensor time series onto lab-measured VO₂/VCO₂ during running, reducing dependence on metabolic carts such as the COSMED K5.
Multimodal biosignal processing across EMG, kinematics, kinetics, and metabolic data to build training data that generalizes across individuals.
At the University Health Network (UHN) / University of Toronto: extending this pipeline to textile-integrated wearable sensors for continuous, out-of-lab monitoring of heart failure patients.

Selected Outputs

J. Choi, J. Kim, H. Kim, et al. — Estimation of Pulmonary Oxygen Uptake (VO2) and Carbon Dioxide Production (VCO2) Using Transfer Learning (revision under review).
Deep Learning-Based Aerobic Capacity Estimation in Running — Senior Researcher, AI Yangjae HUB Seoul Fellow, Ministry of SMEs and Startups, 2023.7–2024.2.
Wearable sensing & textile-based systems for heart failure monitoring — University Health Network (UHN) / University of Toronto, from 2026, supported by the Canada Leads Program (PI: Dr. Darshan Brahmbhatt).