AI ECG Algorithm for Detecting LV Systolic Dysfunction

Left ventricular systolic dysfunction (LVSD) is associated with an increased risk of heart failure, hospitalization, and mortality. Although transthoracic echocardiography is the standard method for assessing left ventricular ejection fraction (LVEF), its widespread use as a screening tool is limited by availability, cost, and the need for specialized personnel. Artificial intelligence (AI)-based electrocardiography (ECG) algorithms have emerged as promising tools for identifying patients with reduced LVEF using routinely acquired ECG signals.

DeepECG LVSD is a deep learning-based ECG algorithm developed to detect LVSD (LVEF ≤40%) from standard 12-lead ECG recordings. Previous retrospective validation studies demonstrated high diagnostic performance; however, prospective clinical validation in real-world practice remains limited.

The purpose of this prospective observational cohort study is to evaluate the diagnostic performance and clinical utility of DeepECG LVSD in adult patients undergoing both ECG and transthoracic echocardiography at Ajou University Hospital. Approximately 15,000 patients aged 19 years or older who have undergone ECG and echocardiography within 30 days will be enrolled.

The primary objective is to assess the accuracy of the AI algorithm for detecting LVSD using echocardiographic LVEF as the reference standard. Diagnostic performance will be evaluated using the area under the receiver operating characteristic curve (AUROC), sensitivity, specificity, positive predictive value, negative predictive value, and overall accuracy.

Secondary objectives include evaluating the association between AI-predicted LVSD and short-term clinical outcomes, including 30-day all-cause mortality, emergency department visits, and heart failure rehospitalization. Exploratory subgroup analyses will assess algorithm performance according to demographic and clinical characteristics, including age, sex, heart failure status, chronic kidney disease, hypertension, diabetes mellitus, and the interval between ECG and echocardiography.

This study is designed as a minimal-risk observational study and will provide prospective evidence regarding the effectiveness of AI-enabled ECG screening for LVSD in routine clinical practice. Findings from this study may support broader implementation of AI-based ECG tools for the early identification of patients at risk for heart failure and reduced left ventricular systolic function.

Inclusion criteria

• Adults aged ≥19 years.
• Patients who underwent both transthoracic echocardiography and 12-lead electrocardiography (ECG) at Ajou University Hospital in the outpatient, inpatient, or emergency department setting.
• ECG and echocardiography performed within 30 days of each other.

Exclusion criteria

• Interval between ECG and echocardiography greater than 30 days.
• Missing or corrupted original ECG waveform data (XML or HL7 format).
• Presence of an implanted cardiac device, including a permanent pacemaker, implantable cardioverter-defibrillator (ICD), or cardiac resynchronization therapy (CRT) device.
• Missing age, sex, or left ventricular ejection fraction (LVEF) data.