AUTOMATED INDEXATION OF AN IMAGE DATABASE AND A RADIOLOGY REFERENCE SERVER - USE FOR REFERENCE SEARCH AND COMPUTER-ASSISTED TRAINING

We indexed the contents of a radiology server on the web to facilitate access to research documents and to link reference texts to images co ntained in radiology databases. Indexation also allows case reports to be transformed with no supplementary work into formats compatible wit h computer-assisted training. Indexation was performed automatically b y ADM-Index, the aim being to identify the medical concepts expressed within each medical text. Two types of texts were indexed: medical ima ging reference books (Edicerf) and case reports with illustrations and captions (Iconocerf). These documents are now available on a web serv er with HTML format for Edicerf and on an Oracle database for Iconocer f. When the user consults a chapter of a book or a case report, the in dexed terms are displayed in the heading; all reference texts and case reports containing the indexed terms can then be called up instantane ously. The user can express his search in natural language. Indexation follows the same process allowing instantaneous recall of all referen ce texts and case reports where the same concept appears in the diagno sis or clinical context. By using the context of the case reports as t he search index, all case reports involving a common medical concept c an be found. The context is interpreted as a question. When the user r esponds to this question, ADM-lndex compares this response with the an swer furnished by the reference texts and case reports. Correct or err onenous responses can thus be identified, converting the system into a computer-assisted training tool.