Charles E. Kahn, Jr., MD, Medical College of Wisconsin; Alex Kogan; David S. Channin, MD

To test the feasibility of augmenting online radiology journal articles with context-sensitive medical information and images.

Online journals offer opportunities to link their articles’ web-based content to external knowledge resources. We created and tested a system to provide on-demand, context-sensitive information retrieval from radiology journal articles.

When a user double-clicked on a word in Hotlight-enabled journal article, the user’s web browser opened a new window that displayed information from the RadLex®, PubMed®, and ARRS GoldMiner® knowledge resources. Matching to RadLex and the Medical Subject Headings (MeSH®) terms was done by finding the longest matching string.

RadLex. The RadLex vocabulary is a set of terms developed by the Radiological Society of North America (RSNA) to index clinical and educational materials in radiology.[1-2] The Hotlight server attempted to match the selected words with one or more RadLex terms. If one term matched exactly, only that term was displayed. Otherwise, the system displayed up to three potential RadLex term(s). RadLex definitions were displayed, if available. Each RadLex term was hyperlinked to its corresponding entry on the RadLex web site. Users could follow that link to browse the RadLex vocabulary and its conceptual hierarchy within a new window.

PubMed. PubMed is the web-based interface into the U.S. National Library of Medicine (NLM) database of biomedical literature, which is categorized and indexed using the MeSH vocabulary.[3] Hotlight displayed up to three matching MeSH terms and their definitions. In the Hotlight display, each MeSH term was hyperlinked to PubMed. Thus, users could click on a term to initiate a literature search of that subject term. By default, the search was limited to diagnostic imaging articles.

ARRS GoldMiner. The GoldMiner Image Gallery is a system for asynchronous retrieval of images from the GoldMiner database.[4] Given any query string, GoldMiner attempts to retrieve relevant images. For Hotlight, we used the RadLex term or terms identified by the server as the query string for GoldMiner. Preliminary analysis has shown that GoldMiner yields very high precision in retrieving relevant images; in its initial evaluation, 96% of retrieved images were relevant.[4]

The line of HTML code correctly activated Hotlight functionality in each of the 12 journal articles in our sample.

Hotlight met our goals for integrating context-sensitive knowledge into radiology journal articles:

(1) The system should require as little modification of the original online journal article as possible. The on-screen presentation and layout of the journal should be unchanged, and the document’s HTML code should be altered as little as possible.

(2) The system should facilitate concept-based retrieval of pertinent information. For example, a user reading an article that mentions “renal cell carcinoma” should be able to instantly view the term's definition, retrieve a gallery of pertinent images, and link to other articles in the biomedical literature.

(3) The system should use open standards and employ widely-used controlled vocabularies.

We have developed a system to augment viewing of radiology journal articles through real-time integration of external information resources. The system, called Hotlight, is activated by inserting a single line of HTML code into an online journal article. When a user highlight text within the article, Hotlight searches for matching controlled-vocabulary concepts. Hotlight displays the concept's definition, presents related images from an image library, and provides links to external information resources, such as PubMed. The system has been tested for implementation into four leading online radiology journals.