SCMR Consensus Contour Data

A multi-center cardiac MRI data set with the most robust manual contours defined by the consensus of 7 independent expert readers from 7 world-class core labs.

About

Cardiac MRI has been recognized as the gold standard for quantification of the left ventricular volumes, mass and functions. However, these were shown by studies conducted within a single center. Manual contours vary between centers due to different training protocols. This study aims to bring a consensus ground truth data for myocardial contours derived from multiple centers worldwide.

Seven independent expert readers, representing seven experienced CMR core labs, have analyzed fifteen CMR studies according to their own standard operating protocols, using their preferred tools and following the SCMR guidelines for post-processing of cardiac MRI. The data set consists of mixed normal and patients with heart failure, myocardial infarction and hypertrophy. Images were acquired by different MR modalities and from different sites.

The Data

Size 15 cases
Pathology 5 healthy volunteers
6 myocardial infarction
2 heart failure
2 hypertrophy
Male/Female 11/4
Age 42--77
Images short-axes cine, long-axes cine MRI
Vendor GE (8), SIEMENS (5), PHILIPS (2)

You can also assess your contours !

... whether you are training a new reader or developing your segmentation algorithm ...

Register

Complete and submit the form below. Your request will be reviewed by the data contributors.

Download

After approval, you will be able to download the data set directly.

Contouring

Start contouring only at ED and ES frames according to the SCMR guideline using any tools you like.

Submit

Submit your contours to us for assessment.

Report

We will send you back the report of your contours validated against the consensus.

Request SCMR Consensus Data

Contouring

No papillary muscles. No trabeculae. Only at specified ED and ES frames. Short-axis MRI. Follow the SCMR standard guidelines.

Contour Format

We prefer a generic text format that contains floating-point numbers of contour points in the image coordinate. However, we are able to read contours from major cardiac MRI analysis tools, such as QMass (Medis, the Netherlands), CMR42 (Circle Cardiovascular Imaging Inc., Canada) and OSIRIX (Pixmeo SARL, Switzerland).

If you develop a software or using other software for contouring and would like us to recognise your contour format, then please contact us.


 PatientID: SCP0001001
 Frame: 1
 Image: IM-0001-0201.dcm
 Contour: endo
 Points: 21
 138.750000 128.250000
 138.750000 128.500000
 138.750000 128.750000
 .
 .
 Contour: epi
 Points: 18
 115.500000 108.000000
 115.750000 107.750000
 .
 .

Reporting

We will send technical and clinical validation of your contours against the consensus. Technical validation includes contour-based distance measurements, the Dice indices, plots of disagreement areas, and volume differences. Clinical validation includes Bland-Altman plots on end-diastolic volumes, end-systolic volumes, mass, ejection fractions, and reader biases compared with the seven readers.

Note that we do not share the consensus contour points to you in order to keep the objectivity of this assessment.

Data Contributors

Eike Nagel
University Hospital Frankfurt, Germany
Alistair Young
University of Auckland, New Zealand

Supported by