DFKI’s MEDICO Semantic Search Demonstrator

I’ve produced a couple of screencasts of the current version of DFKI’s demonstrator for manual semantic annotation of medical images in DICOM format. Annotations are stored in RDF format.

The tool is divided into two main components: An annotation and a retrieval interface.

Annotation Interface #1

Please click on the image to open the screencast in a new window.

The video shows a recent version of the annotation frontend. The loaded DICOM image (currently only 2d is supported) is displayed using ImageJ. There is a schematic drawing of a human which is used to show the body region of the current image.

Below you find the Contrast Windowing component which uses radiodensity annotations for certain tissue types in the MEDICO Ontology to adjust the level and window of the currently loaded image.

The “Toolbar” component is used to choose from a set of different geometric primitives to draw regions of interest (ROIs) in the image. These ROIs can be moved and resized. Currently, we only support rectangles and ovals. We plan to support arbitrary polygons next. The coordinates and parameters of these geometries are saved semantically in the triple store.

After placing a new ROI on the image, it can be added to the list in the ROIManager component and annotated semantically.

Anatomy and disease dimension are annotated separately. The vocabulary is from RadLex (latest version: the FMA fragment of the current body region) and the ICD9. In a newer version of the demonstrator, we also have multi-language support using the synonyms in the FMA (English, German, Spanish, Filipino (!),…). Again, there is search-as-you-type for the comboboxes.

Annotation Interface #2

Please click on the image to open the screencast in a new window.

It shows the annotation of a second image but with different annotations.

Semantic Search Interface

Please click on the image to open the screencast in a new window.

demonstrates the semantic search. None of the images before has been annotated with “thorax” but they are still found and ranked by their “semantic distance”. By clicking on the anatomical annotation “lung zone” of the first hit a browser pops up and displays the respective RadLex page. “Lung zone” is a subclass of “lung” which itself is a subclass of “thorax”. This explains, how it could be found by searching for “thorax”

It is also possible to search for the definition of the ICD9 term by clicking on the respective disease annotation.

The annotations can also be visualized using Prefuse. By clicking on the “Expand” button, the next level in the ontology hierarchy around a concept is expanded and displayed. This is quite handy for an exploration of the ontology.

Finally, by clicking on the image link, the image can be opened.

Tags: , , ,

Leave a Reply