Current Organizations / Institutes

The organizations/institutes from the scholar's latest article


Kyung Su Kim has a total of 214 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    Artificial Intelligence-Enabled Electrocardiography Model for Left Ventricular Systolic and Diastolic Dysfunction Predict Incident Atrial Fibrillation in Patients with Sinus Rhythm: A Time-Dependent Analysis

    by Kyung Su Kim, Jeong Min Son, Hak Seung Lee, Soobeen Seol, Sora Kang, Joon-Myoung Kwon, Min Sung Lee, Kyung-Hee Kim

    Background/Objectives: Artificial intelligence-enabled electrocardiography (AI-ECG) can infer left ventricular systolic dysfunction (LVSD) and diastolic dysfunction (LVDD) from a standard 12-lead tracing. Whether these structural AI-ECG scores predict incident atrial fibrillation (AF) in patients in sinus rhythm, and how their predictive value changes over time, is unclear. Methods: In a retrospective single-center cohort of patients with a sinus-rhythm index ECG, AI-ECG LVSD and LVDD scores were derived and dichotomized at prespecified cutoffs (LVSD ≥ 9.7; LVDD ≥ 20.8). The outcome was incident AF, assessed within 30, 90, 180, 270, and 365 days. Discrimination was quantified by the AUROC. Because the proportional hazards assumption was violated, the time course of risk for the joint LVSD × LVDD classification was characterized with window-specific Cox, Aalen additive hazards, and restricted mean survival time analyses. Results: Among 19,593 index ECGs from 18,984 patients, 1190 (6.07%) were followed by incident AF within one year. One-year incidence rose from 4.1% (both negative) to 9.8% (LVSD-positive only), 20.1% (LVDD-positive only), and 19.9% (both positive). Discrimination was highest for early events and attenuated over time, more steeply for LVSD (AUROC 0.803 at 30 days to 0.713 at 365 days) than LVDD (0.817 to 0.764). Combining AI-ECG LVSD and LVDD models, the excess risk conferred by dual positivity was concentrated in the first weeks after the index ECG and, consistently across cumulative Cox, Aalen additive hazards, and restricted mean survival time analyses, converged with that of isolated LVDD positivity by one year. Conclusions: Structural AI-ECG LVSD and LVDD scores from a single sinus-rhythm ECG predict incident AF in a time-dependent manner, with the strongest performance shortly after acquisition and more durable discrimination for LVDD. A single AI-ECG may help target short-term AF surveillance, particularly in patients with combined systolic–diastolic dysfunction.

  • Open Access

    Training Trajectories Reveal Cross-Modality Generalization Gaps in 12-Lead–Trained AI-ECG Deployed on Smartwatch Recordings (Preprint)

    by Hak Seung Lee, Jong-Hwan Jang, Sora Kang, Yong-Yeon Jo, Jeong Min Son, Min Sung Lee, Kyung Su Kim, Joon-Myoung Kwon, Kyung-Hee Kim

    UNSTRUCTURED AI-ECG models trained on 12-lead–derived Lead I demonstrated comparable endpoint discrimination on smartwatch-derived Lead I but substantially divergent training trajectories, revealing cross-modality generalization gaps not detected by endpoint metrics alone.

  • Non-invasive classification of coronary perfusion pressure during CPR using smartphone-based skin video and deep learning

    by Soyoon Kwon, Dong Ah Shin, Taegyun Kim, Kyung Su Kim, Gil Joon Suh, Jaehoon Sim, Sungmoon Hur, Jaeheung Park, Jung Chan Lee


  • Top Cited Papers
    Open Access

    Sit‐to‐Stand Power From 2D Pose Estimation as an Indicator of Muscle Strength in Older Adults

    by Wonjae Hwang, Dain Shim, Joowan Kim, Kyung Rok Oh, Sun Gun Chung, Jaewon Beom, Myung Woo Park, Kyung Su Kim, Joonghee Kim, Chul Hyun Park, et al.

    Background: The sit‐to‐stand (STS) test is a crucial tool for sarcopenia assessment. However, the two most widely used diagnostic frameworks differ in their conceptualization of the test. The European Working Group on Sarcopenia in Older People 2 views the STS test primarily as a proxy for muscle strength, while the Asian Working Group for Sarcopenia 2019 (AWGS 2019) classifies it as a measure of physical performance. This discrepancy poses challenges to conceptual clarity in both research and practice. Recent advancements in pose‐estimation algorithms allow for kinematic assessment using a standard handheld devices, providing a simple and cost‐effective alternative to conventional motion‐analysis. Methods: A total of 129 community‐dwelling older adults (mean age 71.8 years) who could ambulate independently underwent assessments of muscle strength (knee extensor and handgrip strength), physical performance (usual walking speed, Timed Up and Go [TUG] velocity and Short Physical Performance Battery [SPPB]) and muscle mass (appendicular skeletal muscle index by bioelectrical impedance analysis). Kinematic analysis of the 30‐s sit‐to‐stand (30‐s STS) was conducted using 2D pose estimation to derive peak STS power and joint angles. For validation, a subset of 20 participants completed 30‐s STS trials that were recorded concurrently using 2D pose estimation, optical motion capture and force‐plate recordings. The correlation between peak STS power, five times sit‐to‐stand (5x‐STS) time, and 30‐s STS repetitions with clinical indicators was examined. Results: Peak STS power correlated strongly with muscle strength and mass (knee extensor strength r = 0.64; handgrip r = 0.64; ASMI r = 0.70; all p < 0.001) but only weakly with physical performance (usual walking speed: r = 0.19, p = 0.030; TUG velocity and SPPB: p > 0.05). In contrast, the 5x‐STS time and 30‐s STS repetitions showed moderate to strong associations with physical performance (e.g., SPPB r = −0.79 and 0.52, respectively; both p < 0.001) but did not correlate with strength or muscle mass. Despite systematic biases between pose estimation and the reference, the agreement was high (intraclass correlation coefficient for peak STS power = 0.94; joint‐angles = 0.83–0.89). In sex‐stratified, within‐subject mixed‐effects models, changes in posture and timing explained modest variance in ΔPeak STS power (marginal R 2 = 0.024 in men; 0.138 in women). Conclusions: Peak STS power correlated more strongly with muscular strength, while 5x‐STS and 30‐s STS tests were more closely related to physical performance. These findings align better with and provide stronger support for the AWGS 2019 framework.


  • Open Access

    Minimum required distance for clinically significant measurement of habitual gait speed

    by Myung Woo Park, Sun Gun Chung, Jaewon Beom, Kyung Su Kim, Joonghee Kim, Chul-Hyun Park, Jinkyu Lee, Keewon Kim

    Background: Gait speed indicates morbidity and life expectancy in older adults, but the minimum walking distance for measurement remains unclear. This study aimed to determine the minimum distance required to measure clinically reliable gait speed using a smartphone camera and pose estimation, and whether the distance is influenced by subject characteristics or measurement methods. Methods: Twenty-four healthy volunteers (≥ 65 years old) performed a video-recorded 10-m gait test, including acceleration and deceleration. Fourteen body points were derived using a pose-estimation algorithm. Speed was calculated based on the center of mass or a leading foot which simulates a condition with a walkway sensor and validated against manual measurements. Multiple videos of gait over varying distances were obtained by cropping video frames at 0.1-m intervals. Variance in gait speed over specific distances was calculated using ANOVA. “Minimum required distance” was defined as the shortest distance where the confidence interval of gait speed did not exceed the minimal clinically important difference (0.1 m/sec). We also investigated which clinical, anthropometric, or epidemiological variables might influence it by assessing their association with gait speed variance using mean squared error from linear regression. Results: Gait speed measured by pose estimation (1.55 ± 0.18 m/s) showed a high level of agreement with manual measurement (1.56 ± 0.14 m/s), with an intraclass correlation coefficient of 0.889 (95% CI: 0.822–0.931). “Minimum required distance” was 2.1 m when gait speed was calculated with the center of mass with 95% confidence interval while “minimun required distance” based on a leading foot was 4.7 m with 90% confidence interval. Gait speed itself and muscle strength were positively correlated with gait speed variance (r = 0.250, p = 0.036 for gait speed; r = 0.312, p = 0.008 for knee extension strength; r = 0.230, p = 0.053 for grip strength), whereas other epidemiologic or clinical parameters, and physical performance scales, were not. Conclusions: Clinically reliable measurement of gait speed could be achieved over 2.1 m using a smartphone and pose estimation inducing the center of mass with 95% confidence interval. The weaker the muscle strength or the slower the gait speed, the shorter the distance might be required.


  • Development of artificial intelligence-driven biosignal-sensitive cardiopulmonary resuscitation robot

    by Taegyun Kim, Gil Joon Suh, Kyung Su Kim, Hayoung Kim, Heesu Park, Woon Yong Kwon, Jaeheung Park, Jaehoon Sim, Sungmoon Hur, Jung Chan Lee, et al.


  • Open Access

    A remote-controlled automatic chest compression device capable of moving compression position during CPR: A pilot study in a mannequin and a swine model of cardiac arrest

    by Gil Joon Suh, Taegyun Kim, Kyung Su Kim, Woon Yong Kwon, Hayoung Kim, Heesu Park, Gaonsorae Wang, Jaeheung Park, Sungmoon Hur, Jaehoon Sim, et al.

    Background: Recently, we developed a chest compression device that can move the chest compression position without interruption during CPR and be remotely controlled to minimize rescuer exposure to infectious diseases. The purpose of this study was to compare its performance with conventional mechanical CPR device in a mannequin and a swine model of cardiac arrest. Materials and methods: A prototype of a remote-controlled automatic chest compression device (ROSCER) that can change the chest compression position without interruption during CPR was developed, and its performance was compared with LUCAS 3 in a mannequin and a swine model of cardiac arrest. In a swine model of cardiac arrest, 16 male pigs were randomly assigned into the two groups, ROSCER CPR (n = 8) and LUCAS 3 CPR (n = 8), respectively. During 5 minutes of CPR, hemodynamic parameters including aortic pressure, right atrial pressure, coronary perfusion pressure, common carotid blood flow, and end-tidal carbon dioxide partial pressure were measured. Results: In the compression performance test using a mannequin, compression depth, compression time, decompression time, and plateau time were almost equal between ROSCER and LUCAS 3. In a swine model of cardiac arrest, coronary perfusion pressure showed no difference between the two groups (p = 0.409). Systolic aortic pressure and carotid blood flow were higher in the LUCAS 3 group than in the ROSCER group during 5 minutes of CPR (p < 0.001, p = 0.008, respectively). End-tidal CO2 level of the ROSCER group was initially lower than that of the LUCAS 3 group, but was higher over time (p = 0.022). A Kaplan-Meier survival analysis for ROSC also showed no difference between the two groups (p = 0.46). Conclusion: The prototype of a remote-controlled automated chest compression device can move the chest compression position without interruption during CPR. In a mannequin and a swine model of cardiac arrest, the device showed no inferior performance to a conventional mechanical CPR device.


  • Open Access

    Lack of Fall Prevention Education in Patients Experiencing Fall-Related Hip Fracture: A Qualitative Study

    by Wonjae Hwang, Kyungrok Oh, Min Gyun Kim, Won Bin Kim, Hyun Seung Lee, Sun Gun Chung, Jaewon Beom, Myung Woo Park, Kyung Su Kim, Joonghee Kim, et al.

    Introduction: We aimed to describe patient awareness regarding fall prevention and education, perceived causes of falls, and changes in attitude after experiencing a fall through interviews with older Korean patients who experienced falls with resultant hip fractures. Materials and Methods: We conducted face-to-face semi-structured in-depth interviews with 11 patients who were admitted to Kangbuk Samsung Hospital for hip fractures caused by falls and were referred to the Department of Rehabilitation Medicine for postsurgical rehabilitation between June 2022 and June 2023. The data were analyzed using the phenomenological method developed by Colaizzi. Results: Before hip fracture, none of the patients had received fall prevention education or perceived its necessity; however, they recognized its necessity retrospectively. Participants described the causes of falls as carelessness, actions taken at the time of the fall, environmental factors, and decreased physical function. Most participants believed that falls could be prevented through personal caution and activity restrictions. Some mentioned fall prevention education, exercise, wearing appropriate shoes, environmental adjustments, and seeking assistance from others as methods of preventing future falls. Most patients reported adopting a safety-seeking attitude after experiencing hip fracture. Many patients had negative thoughts such as guilt or thoughts of death, whereas only a few reported increased interest in education and exercise. Conclusions: We observed a lack of fall prevention education, misunderstanding regarding the cause of falls, and negative psychological changes after experiencing hip fractures due to falls in older Korean individuals. Recognizing and managing patient perceptions is crucial for effective fall prevention, requiring both healthcare provider awareness and active participation from patients and caregivers.


  • Open Access

    The 2022 Annual Report on Toxicology Surveillance and Severe Poisoning Cases at Emergency Departments in Korea

    by Eun Sun Lee, Su Jin Kim, Gyu Chong Cho, Mi Jin Lee, Byung Hak So, Kyung Su Kim, Juhyun Song, Sung Woo Lee

    Purpose: This study investigated the actual incidence of acute poisoning in Korea on a nationwide scale, with the aim of laying the groundwork for future initiatives in prevention, strategic antidote distribution, and the development of effective emergency treatment for acute poisoning.Methods: The study analyzed data from 3,038 patients who presented to emergency departments with poisoning-related conditions from June 1, 2022 to December 31, 2022 at 10 sites in nine cities across the country. We extracted data on general characteristics of the poisoning cases, including demographic characteristics (age and gender), place of exposure, reason for poisoning, route of exposure, and the substance involved in the poisoning incident. Age-related patterns in reasons for poisoning, medical outcomes, frequent and primary poisoning substances, and deaths were also analyzed.Results: The population analyzed in our study was predominantly female, with women constituting 54.74% of all cases. Among infants and children, non-intentional poisoning due to general accidents was the most common cause, accounting for 71.43% of cases. Conversely, suicidal poisoning was more prevalent among teenagers and adults over 20. Fifty-two patients died during the study period, with males comprising approximately two-thirds (67.31%) of these fatalities. Pesticides were the most common poisoning substance among those who died, accounting for 55.77% of such cases. Notably, a significant majority of the victims were elderly individuals aged 60 and above.Conclusion: This study holds substantial significance, since it represents the first comprehensive investigation and analysis of the symptoms, treatment, and causes of death due to poisoning in Korea on a national scale. By substantially expanding the range and types of poisonous substances examined, we were able to more precisely identify the characteristics and clinical patterns of poisoning cases nationwide.


  • Open Access

    A quick Sequential Organ Failure Assessment–negative result at triage is associated with low compliance with sepsis bundles: a retrospective analysis of a multicenter prospective registry

    by Heesu Park, Tae Gun Shin, Won Young Kim, You Hwan Jo, Yoon Jung Hwang, Sung-Hyuk Choi, Tae Ho Lim, Kap Su Han, Jonghwan Shin, Gil Joon Suh, et al.

    Objective We investigated the effects of a quick Sequential Organ Failure Assessment (qSOFA)–negative result (qSOFA score <2 points) at triage on the compliance with sepsis bundles among patients with sepsis who presented to the emergency department (ED).Methods Prospective sepsis registry data from 11 urban tertiary hospital EDs between October 2015 and April 2018 were retrospectively reviewed. Patients who met the Third International Consensus Definitions for Sepsis and Septic Shock criteria were included. Primary exposure was defined as a qSOFA score ≥2 points at ED triage. The primary outcome was defined as 3-hour bundle compliance, including lactate measurement, blood culture, broad-spectrum antibiotics administration, and 30 mL/kg crystalloid administration. Multivariate logistic regression analysis to predict 3-hour bundle compliance was performed.Results Among the 2,250 patients enrolled in the registry, 2,087 fulfilled the sepsis criteria. Only 31.4% (656/2,087) of the sepsis patients had qSOFA scores ≥2 points at triage. Patients with qSOFA scores <2 points had lower lactate levels, lower SOFA scores, and a lower 28-day mortality rate. Rates of compliance with lactate measurement (adjusted odds ratio [aOR], 0.47; 95% confidence interval [CI], 0.29–0.75), antibiotics administration (aOR, 0.64; 95% CI, 0.52–0.78), and 30 mL/kg crystalloid administration (aOR, 0.62; 95% CI, 0.49–0.77) within 3 hours from triage were significantly lower in patients with qSOFA scores <2 points. However, the rate of compliance with blood culture within 3 hours from triage (aOR, 1.66; 95% CI, 1.33–2.08) was higher in patients with qSOFA scores <2 points.Conclusion A qSOFA-negative result at ED triage is associated with low compliance with lactate measurement, broad-spectrum antibiotics administration, and 30 mL/kg crystalloid administration within 3 hours in sepsis patients.