Youngsun Kong

Youngsun Kong, Ph.D.

Postdoctoral Research Fellow, Biomedical Engineering

University of Connecticut, Storrs, CT

From Physiological Signals to Objective Human State Assessment

 ·  CV  ·  Google Scholar  ·  GitHub  ·  LinkedIn

44 peer-reviewed papers
(18 first or co-first author)
NIH F32 Principal Investigator
$600K awarded funding
as PI or Co-I

About

I am a biomedical engineer with a foundation in computer science. During my Ph.D. in Biomedical Engineering at the University of Connecticut, I brought signal processing and machine learning to a clinical problem that still lacks reliable objective measures: pain. My doctoral work used physiological signals to study dental pain and pulp status, and this research has since expanded into an NIH F32-supported program. In parallel, in collaboration with the U.S. Navy, I apply the same approaches to human performance under physiological and environmental stress. Across these areas, one question drives my research: how can we turn complex signals from the body into objective measurements we can trust?

Research

I develop computational frameworks that combine signal processing, machine learning, and AI to extract reliable and physiologically meaningful information from complex biosignals. I also design and conduct controlled human experiments to characterize physiological responses and validate these methods. My research spans robust physiological sensing, quantitative signal characterization, objective pain and nociception assessment, and human-state and performance monitoring under physiological and environmental stress. My long-term goal is to develop physiological assessments that account for individual differences and remain reliable across clinical and human-performance settings.

Physiological Signal Characterization

I develop methods to extract sensitive and interpretable information from physiological recordings. This work examines how skin conductance and skin nerve activity change during controlled stimulation, turning raw signals into quantitative measures of autonomic and pain-related responses.

EDA pain features and time-varying skin nerve activity analysis under a thermal grill paradigm
Figure 1. (A) Quantitative electrodermal activity (EDA) features: the derivative phasic feature (dPhEDA) separates stimulus levels more sharply in time than the conventional phasic feature (PhEDA), and a machine-learning pain index tracks stimulus intensity (r ≈ 0.7). (B) Time-varying skin nerve activity (SKNA) features: SKNA is reconstructed from high-sampling-rate ECG using variable frequency complex demodulation (VFCDM), and its time-varying envelope (TVSKNA) is analyzed with nonlinear complexity measures. Both lines of work share a thermal grill paradigm. Adapted from the publications listed below.

Translational Pain Assessment

I use physiological signals to develop objective measures of pain-related responses during dental testing. This work progresses from distinguishing dental stimulation from stress using electrodermal activity (EDA) to combining multiple physiological signals and individual characteristics for more comprehensive pain assessment.

EDA-based dental stimulus detection and multimodal dental pain detection
Figure 2. (A) EDA-based detection of electric pulp testing (EPT) versus stress during dental examination: dPhEDA-derived features such as skin conductance response (SCR) rate differ between conditions, and an artificial neural network (ANN) reaches 76% sensitivity and 87% specificity. (B) Multimodal dental pain detection combining EDA, SKNA, and R–R interval dynamics, with baseline-conditioned encoding, attention-based fusion, and anxiety scores and sex as covariates (80% binary and 60% three-class balanced accuracy). Adapted from the publications listed below; Panel B is from a preprint under review.

Robust Physiological Sensing

I develop methods that help physiological measurements remain useful when recordings contain movement-related interference or other noise. These approaches either suppress contamination to recover useful information or identify unreliable intervals before further analysis.

PPG motion artifact suppression and 1D U-Net artifact detection in EDA
Figure 3. (A) Artifact suppression in photoplethysmography (PPG): accelerometer information and variable frequency complex demodulation remove motion artifacts, enabling accurate heart-rate tracking during exercise. (B) Automatic artifact detection in EDA: a 1D U-Net takes the EDA signal and its spectrogram as input and flags motion- and noise-contaminated intervals in agreement with human annotation. Adapted from the publications listed below.

Human Nociception and Pain Dynamics

I use controlled human experiments to examine how perceived pain and autonomic responses evolve during sustained stimulation. Related machine-learning work tests whether these physiological responses can distinguish stimulation conditions designed to preferentially engage different sensory nerve fibers.

C-fiber response dynamics and fiber-biased EDA response classification
Figure 4. (A) Perceived pain (visual analog scale, VAS) and sympathetic (EDA) responses to 10 s Aδ-fiber and 60 s C-fiber-biased electrocutaneous stimulation at low, medium, and high intensity. (B) Using later time windows to reduce overlap with early Aδ-like responses, XGBoost distinguishes no stimulation, Aδ-fiber, and C-fiber conditions with 87.4% balanced accuracy and an AUROC of 0.948 (area under the receiver operating characteristic curve). Adapted from the publications listed below.

Human Performance Under Challenging Conditions

I investigate how prolonged wakefulness and cold exposure affect human performance, using behavioral measurements, physiological signals, and machine learning. Collaborative engineering projects extend this work toward wearable monitoring and early warning of physiological risk in extreme environments. This line of work has been conducted in collaboration with U.S. Navy researchers.

Sleep deprivation, cold stress, and extreme-environment monitoring studies
Figure 5. (A) Sleep deprivation and task performance: PVT and MATB performance measures, webcam-based behavioral markers, and speech-based detection of cognitive impairment. (B) Reaction times on spatial processing and memory-and-attention tasks, along with autonomic responses, during cold-air exposure in males and females; heart rate variability (HRV) features detect declines in memory and attention. (C) Collaborative work toward extreme-environment monitoring: a wearable circuit platform, water-resilient electrodes, and EDA-based prediction of central nervous system oxygen toxicity (CNSOT). Adapted from the publications listed below.

Funding & Awards

Awarded Funding

Pending / Submitted

Selected Awards

Publications

44 peer-reviewed journal articles (18 first/co-first; 10 corresponding). Key papers are listed under each research theme above; the full list is in my CV and on Google Scholar.

Featured Review

Background

Appointments

Education

Teaching & Service

Teaching

Mentoring

Editorial & Review Service

Outreach

Contact

Department of Biomedical Engineering, University of Connecticut, Storrs, CT

 ·  LinkedIn

↑