For the effective retrieval of clinical information, the elaborate indexing is essential. Two major types of indexing are the human indexing and the automatic or machine indexing. Human indexing shows higher quality but is time consuming, labor-intensive and inconsistent in term assignment activity.

Using the Unified Medical Language System (UMLS) MetaMap program, we mapped the free text from the diagnosis section of radiology reports into UMLS concepts. To improve the precision of UMLS concept indexing by MetaMap, we evaluated the UMLS subset mapping and semantic type filtering methods, determining the best combination for improved precision.

After calculating the candidates from subset combinations, we obtained more enhanced results by semantic-type filtering.

The results may be improved for the complete automation of indexing process.