ImageCLEFmed 2008 Medical Image Retrieval Protocol

Last updated - 22 December 2008, William Hersh

This page contains the details for the ImageCLEF 2008 medical image retrieval task. Thanks to the efforts of Dr. Charles Kahn of the Medical College of Wisconsin, we have been able to procure a new image collection for 2008. This collection consists of images from two radiology journals, along with their captions, article titles, and linkage to Pubmed and the full text. We have been given permission by the Radiology Society of North America (RSNA) to use all of their images in the Goldminer system from their two journals, Radiology and Radiographics, as mounted by Highwire Press. The new collection has a total of JPEG 67,115 images.

In addition to the images, we are also distributing an XML file with the metadata for these images, which consists of the following fields:
Here is an example of the metadata:
    <caption>&quot;Illustration of a neonate at autopsy whose demise was attributed to thymic death. The caption drew attention to the enormous size of the thymus, which is actually normal in appearance. (Reprinted, with permission, from reference 6.)&quot;</caption>
    <title>The right place at the wrong time: historical perspective of the relation of the thymus gland and pediatric radiology</title>
The images are now available for distribution to research groups. Users registered for CLEF should have the password that enables them to access the images at:


The 67,115 images are provided at three levels of resolution:
There are three files of figure captions:


There are 30 topics for ImageCLEFmed 2008, derived from the topics of 2005-2007, with 10 topics in each of three categories:
The topic files are available in several formats:

Data Files

Here is a list of the data files for the collection:

Submitting runs

The format for submitting results will be based on the trec_eval program as follows:
1 1 r99ja45g1x 1 0.567162 GE_gift
1 1 r99ja45g3z 2 0.441542 GE_gift
Several key points for submitted runs are:

Summary of Results

A total of 15 groups submitted 111 runs. The table below shows the best MAP and P@10 obtained for each system.

               Run Type
System Type
0.04      0.17

0.29      0.46
0.08      0.27

0.29      0.43
0.03      0.11
0.22      0.49


Henning Müller, University and Hospitals of Geneva, Geneva, Switzerland Email:
William Hersh, Oregon Health &Science University (OHSU), Portland, OR, USA Email:
Jayashree Kalpathy-Cramer, Oregon Health &Science University (OHSU), Portland, OR, USA Email: