🫀 Comprehensive ECG Datasets Collection

A curated collection of public ECG datasets for machine learning, research, and clinical applications

(local license file unavailable; see change log)

📋 Table of Contents

• 🏥 Clinical ECG Datasets
• 🔬 Research ECG Datasets
• 🏆 Competition ECG Datasets
• 📊 Dataset Comparison
• 🛠️ Tools & Libraries
• 📚 Benchmarks & Papers
• 🚀 Quick Start Guide
• 📖 Dataset Usage Guidelines
• 🤝 Contributing
• 📄 License

---

🏥 Clinical ECG Datasets

Large-Scale Clinical Datasets

Specialized Clinical Datasets

Continuous Monitoring Datasets

---

🔬 Research ECG Datasets

PhysioNet Research Collections

Extended Research Datasets

---

🏆 Competition ECG Datasets

PhysioNet/CinC Challenges

CPSC (China Physiological Signal Challenge)

Other Competition Datasets

---

📊 Dataset Comparison

By Size and Scale

By Clinical Condition

By Data Type and Format

---

🛠️ Tools & Libraries

Data Access and Processing

Visualization and Analysis

AI-Powered ECG Analysis Tools

ECG Foundation Models

ECG Language Models

Machine Learning Frameworks

---

📚 Benchmarks & Papers

Key Survey Papers

Recent High-Impact Papers (2024-2026)

State-of-the-Art Results

PTB-XL Benchmark (Multi-label Classification)

CODE-II Benchmark (66-class Classification)

MIT-BIH Arrhythmia (5-class)

Foundation Model Benchmarks

Structural Heart Disease Detection

Recent Research Trends (2024-2026)

• Foundation Models: Large pre-trained models achieving expert-level performance (ECGFounder, AnyECG, ECG-FM, CardX, ECGFM)
• ECG Language Models: LLM-based ECG interpretation and report generation (CAMEL, GEM, ELF, RhythmBERT, ECG-GPT)
• Cardiac Event Forecasting: Predicting future adverse cardiac outcomes from ECG signals (CAMEL's ECGForecastBench)
• Grounded/Explainable Interpretation: Linking diagnoses to measurable ECG parameters (GEM, VARS)
• Holistic Health Profiling: ECG-based prediction across 1,000+ conditions including non-cardiac diseases (AnyECG)
• Self-supervised Learning: Learning from unlabeled ECG data; BYOL and MAE outperform contrastive approaches (OpenECG)
• Multi-modal Analysis: Combining ECG time series, images, and text (GEM, EchoNext, Heartcare Suite)
• Large-Scale Datasets: HEEDB (10.6M ECGs), CODE-II (2.7M), MIMIC-IV-ECG (800K), Icentia11k
• Open-Source Platforms: Democratizing ECG deep learning for non-experts (ExChanGeAI, ECG-FM)
• Federated Learning: Privacy-preserving ECG analysis across institutions
• Paper ECG Digitization: Automated recovery of signals from paper records (ECGtizer)
• Graph-Based Representations: Novel approaches for heterogeneous ECG signals (VARS)
• Real-Time Analysis: AI-powered platforms for clinical decision support (DeepECG, Qaly)
• FDA-Cleared Tools: Regulatory-approved AI algorithms for clinical use (HeartKey Rhythm)

---

🚀 Quick Start Guide

1. Environment Setup

pip install wfdb pandas numpy matplotlib scipy
pip install torch torchvision  # For deep learning
pip install scikit-learn xgboost  # For traditional ML

2. Loading PTB-XL Dataset

import wfdb
import pandas as pd
import numpy as np

# Load PTB-XL metadata
Y = pd.read_csv('ptbxl_database.csv', index_col='ecg_id')
X = np.array([wfdb.rdsamp(f'records500/{row.filename_lr}')[0] 
              for _, row in Y.iterrows()])

3. Loading MIT-BIH Dataset

import wfdb

# Load a single record
record = wfdb.rdrecord('mitdb/100')
annotation = wfdb.rdann('mitdb/100', 'atr')

signals = record.p_signal
labels = annotation.symbol

4. Loading CODE-II Dataset (2025)

# CODE-II dataset access instructions
# See: https://arxiv.org/abs/2511.15632
# Dataset contains 2.7M ECGs with 66 diagnostic classes
# Access through official CODE-II repository

5. Basic Preprocessing

from scipy import signal

def preprocess_ecg(ecg_signal, fs=500):
    # Bandpass filter (0.5-40 Hz)
    b, a = signal.butter(2, [0.5, 40], btype='band', fs=fs)
    filtered = signal.filtfilt(b, a, ecg_signal)
    
    # Normalize
    normalized = (filtered - np.mean(filtered)) / np.std(filtered)
    return normalized

---

📖 Dataset Usage Guidelines

Citation Requirements

When using these datasets, please cite appropriately:

PTB-XL:

Wagner, P., Strodthoff, N., Bousseljot, R. D., Kreiseler, D., Lunze, F. I., Samek, W., & Schaeffter, T. (2020). 
PTB-XL, a large publicly available electrocardiography dataset. Scientific Data, 7(1), 1-15.

MIT-BIH:

Moody GB, Mark RG. The impact of the MIT-BIH Arrhythmia Database. 
IEEE Eng in Med and Biol 20(3):45-50 (May-June 2001).

CODE-II:

[Citation information to be added - see arXiv:2511.15632]

Icentia11k:

[Citation information to be added - see PhysioNet]

HEEDB:

Reyna, M.A., Deepanshi, Weigle, J., et al. (2026).
The Harvard-Emory ECG Database. Scientific Data.

MIMIC-IV-ECG:

Gow, B., Pollard, T., Nathanson, L.A., Johnson, A., Moody, B., Fernandes, C., et al. (2023).
MIMIC-IV-ECG: Diagnostic Electrocardiogram Matched Subset. PhysioNet.

EchoNext:

[Citation information to be added - see PhysioNet]

Ethical Considerations

• Privacy: All datasets are anonymized, but follow institutional guidelines
• Clinical Use: These datasets are for research only, not clinical diagnosis
• Bias: Be aware of demographic and geographic biases in datasets
• Validation: Always validate models on independent test sets

Data Preprocessing Best Practices

1. Filtering: Apply appropriate bandpass filters (typically 0.5-40 Hz)
2. Normalization: Standardize signals for consistent model training
3. Segmentation: Use appropriate window sizes (typically 2.5-10 seconds)
4. Augmentation: Consider data augmentation for small datasets
5. Quality Control: Remove noisy or corrupted recordings

---

🤝 Contributing

We welcome contributions to this repository! Here's how you can help:

Adding New Datasets

1. Fork this repository
2. Add dataset information to the appropriate table
3. Include proper citations and links
4. Verify all information is accurate
5. Submit a pull request

Required Information for New Datasets

• Dataset name and year
• Number of records and patients
• Data format and specifications
• Access requirements and licensing
• Official links and citations
• Any special features or limitations

Updating Existing Information

• Correction of errors
• Addition of new papers or benchmarks
• Updates to access links
• Performance improvements

Guidelines

• Verify all links are working
• Include proper citations
• Use consistent formatting
• Provide accurate technical specifications

---

🏷️ Tags and Keywords

ecg-datasets electrocardiogram cardiology machine-learning deep-learning foundation-models ecg-language-models arrhythmia heart-rhythm physionet clinical-data medical-ai signal-processing healthcare biomedical-engineering cardiac-monitoring ecg-classification heart-disease medical-datasets public-health cardiovascular wearable-devices

---

📄 License

This repository is licensed under the MIT License. However, individual datasets may have their own licenses - please check each dataset's specific licensing terms before use.

---

📞 Contact & Support

• Issues: GitHub Issues
• Discussions: GitHub Discussions — not enabled

---

⭐ Star History

If you find this repository useful, please consider giving it a star!

Last Updated: July 2026 | Total Datasets: 90+ | Total Records: 20,000,000+