SCR database: Segmentation in Chest Radiographs

Introduction

The automatic segmentation of anatomical structures in chest radiographs is of great importance for computer-aided diagnosis in these images. The SCR database has been established to facilitate comparative studies on segmentation of the lung fields, the heart and the clavicles in standard posterior-anterior chest radiographs.

In the spirit of cooperative scientific progress, we freely share the SCR database and are committed to maintaining a public repository of results of various algorithms on these segmentation tasks. On thes pages, instructions can be found on downloading the database and uploading results, and benchmark results of various methods can be inspected.

Using the database

The data included in this database can be used, free of charge, for research and educational purposes. Redistribution and commercial use is prohibited. Any researcher reporting results which use this database must acknowledge the SCR database. We kindly request you to do so by citing this publication:

In addition, you should also cite the following publication which describes the JSRT database (images were taken from this database):

We appreciate to hear about any publications that use the SCR database. Feedback on the database and this website is also welcome. The person to contact is Bram van Ginneken.

Data description

All chest radiographs are taken from the JSRT database. This is a publicly available database with 247 PA chest radiographs. In each image the lung fields, heart and clavicles have been manually segmented to provide a reference standard.

For more information, see the data description page.

Download

The download page gives instructions on downloading the SCR database.

Results

The result browser allows interactive viewing of images and segmentations and is also a useful way to explore the database.

A tabular overview of the results of various algorithms on each objects can be found here.

Add your results

The data has been subdivided in two folds. Fold 1 consists of all odd numbered images in the JSRT database. Fold 2 consists of all even numbered images. To ensure the integrity of results, you should use the images and reference standard in one fold to train and tune your algorithms, and apply the algorithm to segment the images in the other fold.

We invite other researchers to run their segmentation algorithms against the SCR database. If the results are sent to us, we will then run our evaluation software, and report the algorithm's performance on this site. Currently our evaluation is performed on hard (binary) classification images, which should be supplied in the same format as the binary masks that can be downloaded. Additionally, a literature pointer, article, report or web site that we can refer to for a description of the applied algorithm is requested.

To upload your results, contact Bram van Ginneken by e-mail.

-