UNITY EchoPhase — Multibeat Echocardiographic Phase Detection
Download Dataset
Code Repository
Trained Model
References
Overview
The UNITY EchoPhase dataset provides echocardiographic videos annotated with key cardiac phases — end-systolic (ES) and end-diastolic (ED).
These phases are essential for calculating heart function measures such as ejection fraction, chamber volume, and myocardial strain.
The dataset enables research into automated phase detection and provides a benchmark resource for validating methods against expert human annotations.
We took a large random sample of echocardiographic studies (2010–2020) from Imperial College Healthcare NHS Trust’s echocardiogram database.
- Ethical approval was obtained from the Health Regulatory Agency for the anonymised export of imaging data.
- Patient consent was not required since data had originally been acquired for clinical purposes.
- Acquisition: Images were collected during examinations by experienced echocardiographers, following standard protocols on GE and Philips ultrasound systems.
- Inclusion criteria: Only studies with full demographic data, and without intravenous contrast administration.
- Anonymisation: Automated pipeline removed all patient-identifiable information.
- View selection: A CNN (developed in-house) detected different echo views, isolating apical 4-chamber (A4C).
- Dataset size: 1,000 anonymised videos of varying lengths were randomly selected.
- Annotation process:
- Two accredited cardiologists independently selected ED and ES frames.
- Each was blinded to the other’s choices.
- A custom annotation tool mimicking echo hardware (trackball/arrow navigation) was used.
- This setup replicates clinical workflow for Biplane Simpson’s measurements.
- Resolution: All sequences were downsampled to 112×112 pixels via cubic interpolation.
Each video in the dataset is accompanied by a structured label file:
- Op1 → Operator 1 annotations (used for training/testing the model).
- Op2 → Operator 2 annotations (used for evaluation of observer variability).
- AVIname → The anonymised video filename.
- Number of Frames → Total frame count per video.
- Training/Testing → Indicates the dataset split.
- Beats → Up to 4 beats annotated per file (extended beat labels available on request).
- Missing cells → Indicate either:
- The first frame of interest was an ES frame, or
- The annotator did not select a frame.
Example Label File Preview:

Dataset Summary
Attribute |
UNITY EchoPhase |
Source |
NHS Trust PACS Archives, Imperial College Healthcare |
Ultrasound machines |
Philips iE33 xMATRIX, GE Vivid series |
Size |
1,000 videos (2010–2020) |
Video length |
1–3 heartbeats |
Annotations |
2 independent expert annotations (ED & ES) |
Resolution |
(300–768)×(400–1024), downsampled to 112×112 |
Frame rate |
23–102 fps |
Format |
DICOM → AVI |
Use |
Training & Testing |
Request Dataset Access
The UNITY EchoPhase dataset has been made available for research and benchmarking.
It includes:
- 1,000 anonymised echocardiographic videos
- Frame-level ED and ES labels from two cardiology experts
- Metadata describing operators, beats, and training/testing splits
To request access, please complete the form below:
License
This dataset is licensed under a
Creative Commons BY-NC-ND 4.0 International License.
Project Team
References