10.17037/DATA.00004569
PocketLV is a multi-view POCUS echocardiography dataset with left-ventricular segmentations, covering patients with varied cardiac health, where image quality reflects real-world acquisition by a minimally-trained sonographer.
This dataset contains images slices derived from cine loops and DICOM images collected during point-of-care echocardiography examinations.
Images were collected during the TB-HEART study, a cross-sectional longitudinal study exploring cardiac pathology in adults with pulmonary tuberculosis. All participants provided written informed consent. POCUS echocardiograms were collected using a Clarius HD3 handheld ultrasound device at a primary care facility in Zambia. Scans were performed by an infectious disease clinician, who had completed some training in echocardiography in the United Kingdom (UK).
The data includes apical four-chamber (A4C), parasternal short-axis (PSAX), and subcostal four-chamber (S-A4C) views from 64 patients. In total, among the 64 patients in the dataset, 34 had PSAX views, 27 had S-A4C views, and 22 had A4C views. Cardiac pathology affecting heart function was observed in a subset of patients, such as pericardial effusion, left ventricular systolic dysfunction, and constrictive physiology.
Annotations of the LV myocardium were done using an internal segmentation tool under the guidance of two expert cardiologists to ensure clinical relevance and accuracy.
Here we provide the POCUS scans along with its segmentation masks. We also provide splits used for model training:
Zambia
All segmented images were reviewed with the principal investigator to the TB-HEART study who provided baseline prevalence data of echocardiography defined cardiac pathology outcomes. Segmentation was reviewed by the team creating the dataset.
People with pulmonary tuberculosis (TB) aged 18 or over, who had provided written informed consent (mean age 35.8 ± 10.8 years; 75% male).
All images contained in this dataset are individual slices of echocardiography images randomly assigned to different folders within the dataset, which will be used for benchmarking purposes to train exploratory large visual models.
Initial data collection assigned a unique patient identifier to images collected from each participant guaranteeing anonymisation. Unique patient identifiers in the original dataset are unrelated to the folder enumeration in the PocketLV dataset.
| Organisation | Ethics ID | Other information |
| LSHTM Ethics Online | 28236 | Granted 11 Aug 2023 |
| University of Zambia Biomedical Research Ethics Committee (UNZABREC) | 4108-2023 | Granted 10 July 2023 |
Point-of-care echocardiography; Tuberculosis; Sub-Saharan Africa
English
| Project name | Funder/sponsor | Grant number |
| TB-HEART Studies | Wellcome Trust | 227510/Z/23/Z |
| UKRI AI Centre for Doctoral Training in Digital Healthcare | UK Research and Innovation | EP/Y030974 1 |
| Forename | Surname | Faculty / Dept | Institution | Role |
| Marcello | Scopazzini | Epidemiology and Population Health / Dept of Non-communicable Disease Epidemiology | London School of Hygiene & Tropical Medicine | [1] Contact Person, [2] Data collector and PI to the TB-HEART Study, [3] Data Manager, [7] Project Leader, [12] Researcher, [17] Work Package Leader |
| Lucas | Iijima | Department of Computing | Imperial College London | [12] Researcher, [18] Data Creator |
| Nida | Ruseckaite | Department of Bioengineering | Imperial College London | [12] Researcher, [18] Data Creator |
| Dario | Sesia | National Heart & Lung Institute | Imperial College London | [12] Researcher, [18] Data Creator |
| Amit | Kaura | National Heart & Lung Institute | Imperial College London | [12] Researcher, [18] Data Creator |
| Jamil | Mayet | National Heart & Lung Institute | Imperial College London | [16] Supervisor |
| Anoop | Shah | Epidemiology and Population Health / Dept of Non-communicable Disease Epidemiology | London School of Hygiene & Tropical Medicine | [1] Contact Person, [12] Researcher, [16] Supervisor |
| Choon Hwai | Yap | Department of Bioengineering | Imperial College London | [16] Supervisor |
| Filename | Description | Access status | Licence |
| Myocardium_data | The Myocardium folder contains 3 sub-folders - A4C, PSAX, and Subcostal. Each subfolder contains images and labels (segmentation masks) in the original size of 1024x768. Images are saved in PNG format. | Request access for all | Data Sharing Agreement |
| 4559_User Guide | User guide for dataset (this document) | Open access | Creative Commons Attribution (CCBY) |
The Myocardium folder contains 3 sub-folders - A4C, PSAX, and Subcostal. Each subfolder contains images and labels (segmentation masks) in the original size of 1024x768. Images are saved in PNG format.
Three text files are also provided: