ECG-based transfer learning for cardiovascular disease: A scoping review
An electrocardiogram (ECG) measures the electrical activity of the heart. It is valuable for the early detection of cardiovascular disease (CVD). Advancements in medical diagnosis using artificial intelligence have popularised transfer learning among researchers in machine learning and deep learning...
Saved in:
| Main Authors: | , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
KeAi Communications Co., Ltd.
2025-12-01
|
| Series: | International Journal of Cognitive Computing in Engineering |
| Subjects: | |
| Online Access: | http://www.sciencedirect.com/science/article/pii/S2666307425000075 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | An electrocardiogram (ECG) measures the electrical activity of the heart. It is valuable for the early detection of cardiovascular disease (CVD). Advancements in medical diagnosis using artificial intelligence have popularised transfer learning among researchers in machine learning and deep learning. This scoping review focuses on the application of ECG-based transfer learning for CVD identification, examining the most common types of CVD studied, the input formats used, frequently referenced databases, and the extent of transfer learning's application in diagnosing CVD through ECG analysis. We also conducted a bibliographic analysis to provide an overview of the current situation in the CVD research field. Our analysis of over 70 studies published in the last six years indicates that arrhythmias are the most extensively studied CVD, with the MIT-BIH arrhythmia dataset being the most commonly used in previous research. Pre-trained models such as ResNet, AlexNet, and VGG, which are trained on ImageNet, are often used with two-dimensional (2D) ECG data as network input, achieving accuracy rates exceeding 90 %. The bibliographic analysis indicates that China and India are the highest contributors in this field. Our study also identifies several issues requiring further investigation and recommends areas for future research, including hyperparameter tuning to enhance model performance. |
|---|---|
| ISSN: | 2666-3074 |