Robert Patton PhD, Presenter: Nothing to Disclose

Thomas E Potok PHD, Abstract Co-Author: Nothing to Disclose

Georgia D. Tourassi PhD, Abstract Co-Author: Nothing to Disclose

Barbara Beckerman MBA, Abstract Co-Author: Nothing to Disclose

Christopher Stahl, Abstract Co-Author: Nothing to Disclose

The ability to have a keen awareness and availability of relevant information provides a critical competitive edge in Radiology in which medical imaging technology and applications advance at a rapid pace. Unfortunately, as more web-based Radiology resources become available, identifying and navigating to relevant information becomes a significant challenge. We describe a system that uses a set of significant documents as seed documents (instead of traditional ontology-based approaches) to recommend relevant new documents across various Internet channels. This enables 1) discovery of new internet channels that may be of interest, and 2) a refinement of the information within channels to only that which is most relevant or interesting to the user/radiologist.

The Internet contains an enormous amount of data that streams faster than can be humanly processed and analyzed. In order for Radiology researchers and practitioners to leverage this data, a recommender system was designed and developed. This system helps fill a critical gap that exists in current technology that can enhance awareness in their respective field or topic ofinterest.

Our exhibit will allow the user/radiologist to test this powerful platform to streamline the Internet navigation process in search of relevant textual information.

Our system is called Distribute The Highest Selected Textual Information (DTHSTR). DTHSTR has three components: ingest, analysis, and output. The first component monitors a list of RSS feeds and information channels (e.g., web-based sources such as online journals, news sites, and government sites of interest to radiologists) determined by the user. The user also provides examples of significant documents to be used as reference points for searching related Internet content. The second system component applies a vector space model using the term frequency / inverse corpus frequency term weighting scheme and a cosine similarity measure to identify relevant documents similar to the provided examples. The third system component provides the recommendation results via RSS feeds according to categories (e.g., journal publications, online news, funding opportunities, etc.).

Patton, R, Potok, T, Tourassi, G, Beckerman, B, Stahl, C, A Recommender System for Web-based Discovery and Refinement of Information Radiologists Seek.  Radiological Society of North America 2012 Scientific Assembly and Annual Meeting, November 25 - November 30, 2012 ,Chicago IL. http://archive.rsna.org/2012/12030896.html