Indexing of radiological reports: Moving the human-computer language barrier

Purpose: When using a computerised index of words originating from a large collection of radiological reports, users tend to encounter a human-computer language barrier. This obstacle arises whenever a complex query cannot be adequately formulated in the language required by the computer interface. This study seeks to alleviate this interface problem. Methods and Materials: Since its conception in 2004, the indexing system developed at the Ghent University Hospital covers 130 million words originating from 2 million reports. The index contains 350,000 unique words and results are typically returned within 31 ms. To counter the problems related to complex queries, we educated the users regarding basic search engine features: word grouping, wildcards, and binary logic. We also implemented word stemming in four European languages, automatic rewriting of queries for common synonyms, and contextual searching. Results: Following education, users started to pick up search engine features: phrases (3.0%), wildcards (2.9%), binary operator “and” (4.6%), “not” (1.7%), “or” (1.2%), and parentheses (0.6%). Both the word stemming and the automatic inclusion of synonyms received positive feedback. However, only administrators used contextual searching because of the composite syntax. Conclusion: The search engine features we implemented helped to move the interface language barrier away from our users. These 45 radiologists and researchers use the integrated indexing system on a daily basis as a convenient way of quickly finding radiological studies