.png){kind=icon}
.png){kind=icon}
Return
MECKA: A Multimodal ECG Classification Framework via Knowledge Augmentation
.png){kind=icon}
PRE
AI
DOI：10.1016/j.bspc.2026.110738.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Electrocardiogram (ECG) classification plays a vital role in the early diagnosis of cardiovascular diseases. While recent deep learning (DL) methods enable automated feature extraction, most fail to effectively integrate multi-scale ECG information and incorporate clinical expertise. To address these limitations, we propose MECKA (Multimodal ECG Classification with Knowledge Augmentation), a unified framework that fuses raw ECG signals, temporal–spectral features, and domain-specific textual knowledge. Specifically, MECKA incorporates three key components: (1) a dual-stream fusion module that combines signal- and feature-level information; (2) a text embedding module utilizing large language models (LLMs) trained on curated clinical corpora; and (3) a cross-modal alignment module that aligns ECG and textual representations through contrastive learning. To evaluate MECKA, we conducted extensive experiments on public benchmarks. The results show that our framework achieves state-of-the-art performance, with overall accuracies of 97.73%, 99.82%, and 95.92% on the MultiDB, LTSTDB, and LTAFDB datasets, respectively, while also exhibiting superior efficiency, robustness, and stability compared to existing methods.
Journal
IF:
4.9
Papers:
9.8K
Citations:
2.4W