-
Original Article
-
Korean J Intern Med. 2026;41(1):118-130. Published online January 1, 2026.
DOI: https://doi.org/10.3904/kjim.2025.104
- Enhanced prediction of left ventricular ejection fraction using electrocardiography with the addition of clinical metadata
-
Hyun Woong Park1,2, Taeseen Kang2, Young-Hoon Seo3, Jae-Hyeong Park1
-
1Department of Cardiology, Chungnam National University Hospital, School of Medicine, Chungnam National University, Daejeon, Korea
2Department of Ophthalmology, Chungnam National University Sejong Hospital, Sejong, Korea
3Department of Cardiology, Konyang University Hospital, Daejeon, Korea
- Corresponding author: Jae-Hyeong Park ,Tel: +82-42-280-8237, Fax: +82-42-280-8238, Email: jaehpark@cnu.ac.kr
- Received: March 29, 2025; Revised: July 13, 2025 Accepted: July 25, 2025.
- Abstract
- Background/Aims
Left ventricular ejection fraction (LVEF) is a key echocardiographic parameter for assessing LV systolic function, guiding the management of many cardiovascular diseases, including heart failure (HF). While traditional electrocardiography (ECG) has been widely used in clinical practice, it has limitations in predicting LVEF. This study investigated the impact of integrating ECG data with metadata, such as age, N-terminal pro B-type natriuretic peptide (NT-proBNP), and sodium levels, to enhance the accuracy of LVEF prediction, especially in HF with reduced ejection fraction (HFrEF, LVEF ≤ 40%).
Methods
This retrospective study analyzed ECG and metadata from two tertiary teaching hospitals in Korea. A deep neural network (EfficientNet B3) was trained to predict LVEF, incorporating clinical metadata alongside ECG inputs. Model performance was assessed using the area under the curve (AUC) and the coefficient of determination (R2).
Results
The artificial intelligence (AI) model achieved an AUC of 0.95 when ECG data were combined with age, NT-proBNP, and sodium levels, outperforming models relying on ECG alone (AUC = 0.90). The integration of metadata significantly improved the prediction accuracy, particularly for HFrEF cases. The specificity of the model remained high (96.9%), but sensitivity was relatively low (54.8%), indicating its potential as a screening tool for HFrEF.
Conclusions
The combination of ECG and metadata results using AI enhances the predictive accuracy of HFrEF detection. This approach offers a scalable and noninvasive method for HF screening and risk stratification, particularly in resource-limited settings. Further validation in diverse populations is needed to confirm its clinical utility.
Keywords :Heart failure, systolic; Electrocardiography; Artificial intelligence; Heart failure
