MIMIC-III-Ext-PPG, a PPG-based Benchmark Dataset for Cardiovascular and Respiratory Signal Analysis.

Area of Science:

• Biomedical Engineering
• Medical Informatics

Background:

• Photoplethysmography (PPG) is crucial for non-invasive cardiovascular monitoring.
• Existing PPG datasets often lack scale, quality assessment, or comprehensive annotations.
• The MIMIC-III database provides rich clinical data but requires specialized waveform processing.

Purpose of the Study:

• To introduce MIMIC-III-Ext-PPG, a large-scale, quality-assessed PPG dataset.
• To provide annotations for cardiovascular and respiratory analyses, including heart rhythm, blood pressure, and respiratory rate.
• To establish a benchmarking resource for machine learning in healthcare.

Main Methods:

• Derived PPG segments from the MIMIC-III matched waveform subset.
• Extracted annotations for heart rhythm, blood pressure, respiratory rate, and heart rate from available signals (ECG, RESP, ABP).
• Performed rigorous signal quality assessments for all provided signals.

Main Results:

• MIMIC-III-Ext-PPG contains 6.3 million 30-second PPG segments from 6,189 subjects.
• It is the largest publicly available resource for heart rhythm classification.
• Annotations for systolic/diastolic blood pressure, respiratory rate, and heart rate are provided where applicable.

Conclusions:

• MIMIC-III-Ext-PPG offers an unprecedented, high-quality resource for PPG research.
• The dataset facilitates the development and benchmarking of machine learning models for diverse clinical prediction tasks.
• The dataset is extendable with existing MIMIC-III clinical metadata.