The Participation of MedGIFT Group at ImageCLEFmed 2010

This article presents the participation of the MedGIFT group in ImageCLEFmed 2010. Since 2004, the group has participated in the medical image retrieval tasks of ImageCLEF (ImageCLEFmed) each year. The main goal is to provide a baseline by using the same technology each year, and to search for further improvement if possible. There are three types of tasks for ImageCLEFmed 2010: modality classification, image–based retrieval and case–based retrieval. The MedGIFT group participated in all three tasks. For ad–hoc retrieval and case–based retrieval tasks, two existing retrieval engines were used: the GNU Image Finding Tool (GIFT) for visual retrieval and Apache Lucene for text. Fusion strategies were also tried out to combine results from two engines. For the modality classification, a purely visual approach was used with GIFT for the visual retrieval and a kNN (k–Nearest Neighbors) classifier for the classification. Results show that the best textual run outperforms the best visual run by a factor of 30 in terms of mean average precision. Baselines provided by Apache Lucene and GIFT are ranked above the average among textual runs and visual runs respectively in ad–hoc retrieval. In the case– based retrieval task the Lucene baseline is the third best automatic run. For modality classification, GIFT and the kNN–based approach perform slightly better than the average of visual approaches.