Overview of the ImageCLEFmed 2006 Medical Retrieval and Medical Annotation Tasks

Abstract

Abstract. This paper describes the medical image retrieval and medical image annotation tasks of ImageCLEF 2007. Separate sections describe each of the two tasks, with the participation and an evaluation of major findings from the results of each given. A total of 13 groups participated in the medical retrieval task and 10 in the medical annotation task. The medical retrieval task added two new data sets for a total of over 66'000 images. Topics were derived from a log file of the Pubmed biomedical literature search system, creating realistic information needs with a clear user model. The medical annotation task was in 2007 organized in a new format as a hierarchical classification had to be performed and classification could be stopped at any hierarchy level. This required algorithms to change significantly and to integrate a confidence level into their decisions to be able to judge where to stop classification to avoid making mistakes in the hierarchy. Scoring took into account errors and unclassified parts.

Image Retrieval and Annotation Using Maximum Entropy

“…The results of the evaluation are given in detail in the overview paper [6]. Table 3 gives an overview on our submissions and the best competing runs and it can be seen that the runs using the discriminative classifier for the histograms clearly outperform the image distortion model and that in summary our method performed very good on the task.…”

“…We submitted nine runs to the medical retrieval task [6], one of these using only text, three using only visual information, and five using visual and textual information. For one of the combined runs we used the above-described maximum entropy training method.…”

“…We submitted three runs to the medical automatic annotation task [6]: one run using the image distortion model RWTHi6-IDM, with exactly the same settings as the according run from last year, which clearly outperformed all competing methods [16] and two other runs based on sparse histograms of image patches [14], where we used a discriminatively trained log-linear maximum entropy model (RWTHi6-SHME) and support vector machines with a histogram intersection kernel (RWTHi6-SHSVM) respectively. Due to time constraints we were unable to submit the method described in Section 3.3, but we give comparison results here.…”

Overview of the CLEF 2009 Medical Image Annotation Track

“…It was born as a 60 plain class problem [3], grew up to a 120 class problem [6], and became a complex hierarchical class task in 2007 [5,2]. In 2008, class imbalance was added to foster the use of prior knowledge encoded into the hierarchy of classes [1].…”

Baseline Results for the ImageCLEF 2006 Medical Automatic Annotation Task

“…The ImageCLEF 2006 annotation task [2] consists of 10,000 reference images grouped into 116 categories and 1,000 images to be automatically categorized. This paper aims at providing a baseline for comparison of the experiments in 2005 and 2006 rather than presenting an optimal classifier.…”