Curated Physiological Datasets on PhysioNet

Posted on In , Word count in article: 800 Reading time ≈ 3 mins.
An overview of openly accessible physiological signal repositories from PhysioNet, serving as essential resources for training advanced models and validating biomedical algorithms.

Access Note: All datasets described below have been mirrored for easier access at the LIP6 Nuage Repository.

1. ECG & PPG Signal with Arrhythmia Episodes (2022)

Source: doi.org/10.13026/s32e-sv15

Overview

This resource provides a dedicated software tool for generating synthetic ECG and PPG signals populated with a broad range of arrhythmic events. It is designed to augment training data for arrhythmia detection algorithms.

Key Features

  • Arrhythmia Types: Normal sinus rhythm, Atrial Fibrillation (AF), Bradycardia, Ventricular Tachycardia (VT), and Atrial Premature Beats (APB).
  • Customization:
    • Sampling Frequency: 75–1000 Hz for PPG, 250–1000 Hz for ECG.
    • Signal Parameters: Configurable duration, timing of abnormal episodes, and noise characteristics.
  • Realism: Supports the superimposition of realistic measurement noise to simulate authentic acquisition environments.

2. Motion Artifact Contaminated fNIRS and EEG (2014)

Source: physionet.org/content/motion-artifact/1.0.0

Overview

This dataset contains simultaneous fNIRS and EEG recordings designed to study motion artifact removal. Data was collected in an experimental setting with controlled motion artifacts.

Acquisition Setup

  • fNIRS: ~25 Hz sampling, dual wavelengths (690 nm and 830 nm).
  • EEG: 2048 Hz sampling.
  • Motion Reference: Triaxial accelerometer sampled at 200 Hz.
  • Protocol: One sensor group was deliberately moved to induce artifacts, while a second control group remained stationary.

Data Structure

The dataset includes synchronized fNIRS, EEG, and accelerometer streams.

fNIRS Data Format:

Column Description
1 Sample number
2-3 Raw Light Intensity (690/830 nm) - Channel 1 (25 Hz)
4-5 Raw Light Intensity (690/830 nm) - Channel 2 (25 Hz)
6 Trigger Signal (Rise=Start, Low=Motion, High=Clean)
7-12 Accelerometer Data (Sensor 1 & 2 X/Y/Z)
13 Accelerometer Trigger

EEG Data Format:

Column Description
1 Sample number
2 Raw EEG - Channel 1 (2048 Hz)
3 Raw EEG - Channel 2 (2048 Hz)
4 Trigger Signal
5-10 Accelerometer Data
11 Accelerometer Trigger

Note: Channel 1 is generally the stationary (clean) reference, while Channel 2 contains induced motion artifacts.

3. ScientISST MOVE (2024)

Source: doi.org/10.13026/hyxq-r919

Overview

ScientISST MOVE provides multimodal biosignal recordings captured in natural living environments. It features 17 participants performing annotated daily activities such as walking, running, and social interactions.

Device Configuration

Signal Type ScientISST (Chest/Forearm) Empatica E4 (Wrist)
ECG 500 Hz (Gel electrodes) N/A
PPG 500 Hz 64 Hz
EDA 500 Hz 4 Hz
EMG 500 Hz N/A
ACC 500 Hz 32 Hz
Temp N/A 4 Hz

4. BIG IDEAs Glycemic Variability (2023)

Source: physionet.org/content/big-ideas-glycemic-wearable

Overview

This dataset explores the relationship between non-invasive physiological signals and glycemic variability. It combines continuous glucose monitoring (CGM) with wearable data from 16 participants.

Features

  • Physiological Data: Heart rate, accelerometry, Blood Volume Pulse (PPG), EDA, and skin temperature from Apple Watch and Empatica E4.
  • Glucose Data: Measurements every 5 minutes.
  • Nutritional Logs: Detailed intake records (calories, macros) in Food_Log_xxx.csv.
  • Application: Ideal for researching non-invasive glucose estimation using time-aligned PPG and metabolic data.

5. Labeled Raw Accelerometry Data (2021)

Source: doi.org/10.13026/51h0-a262

Overview

A human activity recognition (HAR) dataset featuring high-frequency accelerometry from 32 healthy adults (13 males, 19 females).

Protocol

  • Sensors: 4x ActiGraph GT3X+ (Left Wrist, Left Hip, Left Ankle, Right Ankle).
  • Sampling: 100 Hz.
  • Activities: Walking (~1 km), Stair Climbing (Up/Down 6x), Driving (~12.8 miles).

Format

Each file contains time-series data with activity codes:

  • 1: Walk
  • 2: Downstairs
  • 3: Upstairs
  • 4: Drive
  • 77: Clap (Sync)
  • 99: Off-protocol

6. Stress and Structured Exercise Sessions (2025)

Source: physionet.org/content/wearable-device-dataset

Overview

A multimodal dataset recording physiological responses to induced acute stress and structured physical exercise using Empatica E4 wristbands.

Protocols

  1. Acute Stress (STRESS): Mental arithmetic and emotional stimulation tasks ($n=36$).
  2. Aerobic Exercise: Moderate, rhythmic cycling ($n=30$).
  3. Anaerobic Exercise: Short, high-intensity cycling ($n=31$).

File Structure

  • BVP.csv: 64 Hz signal for HRV and waveform quality analysis.
  • EDA.csv: Electrodermal activity.
  • TEMP.csv: Skin temperature.
  • tags.csv: Event markers.

7. BigIdeasLab_STEP (2021)

Source: physionet.org/content/bigideaslab-step-hr-smartwatch

Overview

This study assesses the impact of skin tone (Fitzpatrick scale 1–6), activity type, and device model on the accuracy of optical heart rate monitoring.

Demographics & Protocol

  • Participants: 53 individuals (ages 18–54), balanced across skin tones.
  • Activities: Rest, Paced Breathing, Brisk Walking, Typing.
  • Devices: Apple Watch 4, Empatica E4, Fitbit Charge 2, Garmin Vivosmart 3, Xiaomi Miband 3, Biovotion Everion.
  • Reference: Bittium Faros 180 ECG (~1000 Hz).

Note: This dataset provides processed Heart Rate (BPM) values, not raw PPG waveforms.

Summary

The majority of these datasets utilize the Empatica E4 for acquiring PPG, EDA, and temperature signals, establishing it as a common reference in the field. However, researchers should note that the E4 has notably been succeeded by the EmbracePlus, which offers enhanced energy efficiency and multi-channel capabilities.