The Structuring Method of Diagnostic Reports by Using Description Units for Semantic Interpretation
| Autor: | Ando, Yutaka, Kawaguchi, Osamu, Futami, Hikaru, Yamagishi, Hirotada, Fujii, Hirofumi, Tsukamoto, Nobuhiro, Kasamatsu, Tomotaka, Kaneko, Hiroshi, Osada, Masakazu, Kurosaki, Kaori, Kubo, Atsushi |
|---|---|
| Jazyk: | angličtina |
| Rok vydání: | 2007 |
| Popis: | PURPOSE It is useful for references and making diagnosis to search similar case using diagnostic reports from past diagnostic reports that was electronically stored. However, it is difficult to systematically and semantically interpret the descriptive contents of the reports which are generally expressed using free-text format. At first, we have constructed the conversion method from free text format to semantic structure and next the system to create structured reports with semantic structure 'description unit' that we defined from free-text reports. In this study, we proposed the reference system to search similar diagnostic reports using semantic sentence. \nMETHOD and MATERIALS Structured diagnostic reports were converted into the description units which were classified into two types of finding units and diagnostic units. The finding and diagnostic units included finding/diagnosis, modifier, region, regional modifier, and confidence. Diagnostic reports in free-text format were analyzed and classified into the description units by using a text mining method. To search similar case, we developed the retrieval system with the following algorithms: (1) to convert description unites from findings or impressions that physicians wrote, (2) to find past stored reports according the higher match rate. The match rate was defined as the identical ratio of number of description units extracted from findings or impressions, to number of it from past reports. In this study, we assessed the usefulness of this system by applying to clinical diagnostic reports of chest CT studies. \nRESULTS The accuracy of our conversion method was 90% at recall rate. The system could find over 60% of the similar reports that physician requested from 1,500 cases of past reports. \nCONCLUSION Our newly developed system could search similar cases using semantic analysis. In addition, it was expected to search similar images by free text findings. RSNA'07 93th Scientific Assembly and Annual Meeting |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|