HeartBERT: A self-supervised ECG embedding model for efficient and effective medical signal analysis

2026-04-27

PRE

Saedeh Tahery *

Fatemeh Hamid Akhlaghi

Termeh Amirsoleimani

Saeed Farzi

DOI：10.1016/j.compbiomed.2026.111712

Original

Original request for help

Save

Abstract

En 中文

• Bridges the gap between Natural Language Processing (NLP) and healthcare by drawing inspiration from the periodic time series patterns of ECG and natural text. • Proposes a novel quantization method to translate ECG signals into synthetic language representations. • Introduces HeartBERT, a versatile self-supervised model that learns contextual embeddings from unlabeled ECG signals. • Demonstrates the effectiveness of HeartBERT in downstream tasks, including sleep-stage and heartbeat classification, with various transformer layer update strategies. • Analyzes the limitations of ChatGPT (GPT-4o) in downstream tasks, indicating the challenges of domain-specific tasks.

Keywords:

ECG signal processing

natural language processing

self-supervised learning

HeartBERT

medical signal analysis

HeartBERT: A self-supervised ECG embedding model for efficient and effective medical signal analysis

Abstract

Journal

Organization