MIMIC-III-Ext-PPG — ICU Wearable PPG Benchmark [Large-Scale, WFDB Format]

Overview

MIMIC-III-Ext-PPG is a curated, large-scale benchmark dataset derived from the MIMIC-III ICU database, specifically structured for photoplethysmography (PPG) research relevant to wearable health monitoring. It provides quality-assessed 30-second waveform segments including PPG, ECG, arterial blood pressure (ABP), and respiratory signals (RESP).

The dataset addresses a key gap in the wearable sensing research community: prior public PPG datasets were either small in scale or lacked support for multiple classification and regression tasks. MIMIC-III-Ext-PPG enables direct benchmarking of wearable-grade signal processing algorithms for tasks including arrhythmia detection, blood pressure estimation, respiratory rate estimation, and signal quality assessment.

Structured in the WFDB (WaveForm DataBase) format, it is directly compatible with PhysioNet's toolchain and major biomedical signal processing libraries in Python, MATLAB, and Julia.

Column Schema

Key Statistics

• Total Records: large-scale multi-patient ICU dataset (thousands of patients)
• Segment Length: 30 seconds per waveform segment
• Channels: PPG + ECG, ABP, RESP where available
• File Format: WFDB (compatible with Python wfdb, biosppy, neurokit2, MATLAB)
• Source: MIMIC-III clinical database
• Published: February 2026

Use Cases

• Wearable PPG algorithm development and benchmarking for cardiovascular monitoring
• Non-invasive blood pressure estimation from PPG signals
• Arrhythmia detection and respiratory rate estimation using wearable-grade signals
• Signal quality assessment model training for IoT health devices

Source & Attribution

MIMIC-III-Ext-PPG is hosted on PhysioNet and was developed to provide a high-quality, multi-task PPG benchmark for the wearable and digital health research community. It builds on the MIMIC-III clinical database from the MIT Laboratory for Computational Physiology.