Annotation of Chest Radiology Reports for Indexing and Retrieval

摘要

Annotation of MEDLINE citations with controlled vocabulary terms improves the quality of retrieval results. Due to variety in descriptions of similar clinical phenomena and abundance of negation and uncertainty, annotation of clinical radiology reports for subsequent indexing and retrieval with a search engine is even more important. Provided with an opportunity to add about 4,000 radiology reports to collections indexed with NLM image retrieval engine Open-i, we needed to assure good retrieval quality. To accomplish this, we explored automatic and manual approaches to annotation, as well as developed a small controlled vocabulary of chest x-ray indexing terms and guidelines for manual annotation. Manual annotation captured the most salient findings in the reports and normalized the sparse distinct descriptions of similar findings to one controlled vocabulary term. This paper presents the vocabulary and the manual annotation process, as well as an evaluation of the automatic annotation of the reports.