Nishimoto Naoki, Presenter: Nothing to Disclose

Satoshi Terae MD, Abstract Co-Author: Research grant, J-MAC SYSTEM, Inc Research grant, Medical Image Lab Research grant, FUJIFILM Holdings Corporation

Masahito Uesugi, Abstract Co-Author: Nothing to Disclose

Ayako Yagahara, Abstract Co-Author: Nothing to Disclose

Katsuhiko Ogasawara PhD, Abstract Co-Author: Nothing to Disclose

Tsunetaro Sakurai, Abstract Co-Author: Nothing to Disclose

RadLex is terminology used to identify the meaning of medical terms in radiology reports. However, this is still a challenge to map the conceptual label to the specific terms of similar expressions, such as S1, S2 or S3. One is required to use an interface terminology to automatically identify the concept of clinical terms written by a radiologist. The purpose of this study is to create a collection of co-occurred terms with the Couinaud hepatic segment expression and to apply the collection to the concept identification in Japanese CT reports.

We selected 102 Japanese CT reports randomly from 1,989 reports that were made during July 2005 in the Hokkaido University Hospital, Japan. We manually collected the terms that co-occurred with a Couinaud hepatic segment and a lung segment in Japanese CT reports. The co-occurred word collection includes 128 words for the lung segments and 120 for the hepatic segments. To identify the Couinaud hepatic segments and distinguish the lung segments automatically, (both are expressed as “S with number”), an original Java program was developed. We used Protégé 3.3.1 to view and search RadLex. When the Java program detected a “S with number“, it searched the co-occurred terms in the same sentence. If the co-occurred terms for the hepatic segment collection were detected, it mapped the RadLex concept label and ID (e.g. RID60). Accuracy was applied to measure the performance of our application.

The number of characters in each report was 792±273 and the number of character types was 719. The accuracy of “S with number” detection was 86.9%. The overall accuracy was 70.79% for RadLex concept mapping to hepatic and lung segments. The accuracy of hepatic segment mapping was 86.5%. One can use a general parser to identify the hepatic segment expression. Pakhomov et. al. reported that the accuracy of a general parser for the clinical part-of-speech tagging varied from 74.7-92.6%.

Although our approach is limited to the hepatic segment expressions and their concepts, it is practical to use the collection of co-occurred terms to map the RadLex concept label. We showed even the small 128 words set of co-occurred word collection contributed to its accuracy.

Applying reference terminology, RadLex, to information retrieval will contribute to clinical decision support for radiologists.

Naoki, N, Terae, S, Uesugi, M, Yagahara, A, Ogasawara, K, Sakurai, T, Hepatic Segment Phrase Disambiguation Using RadLex.  Radiological Society of North America 2008 Scientific Assembly and Annual Meeting, February 18 - February 20, 2008 ,Chicago IL. http://archive.rsna.org/2008/6019534.html