Structuring of Free Text Diagnostic Report

Background:There is a trend towards digitization of reporting systems, which are mainly used in the research and creation of diagnostic reports. However, the current situation is not at a satisfactory level in terms of effective use of digital information. The main reason is considered to be related to diagnostic reports, which are generally in free text format, thus making it difficult to utilize the descriptive contents systematically. \nPurpose:We have developed a method for automatically creating structured reports in free text format. This method uses a text mining tool that extracts semantic information from the contents in free text format to enable semantic interpretation of the information. \nResult:The match rate is 70.2% when the number of learning reports is 100, it rises to 85.5% when the number of learning reports is increased to 500. In addition, when the number of learning reports is about 300 or more, the match rate tends to saturate and becomes nearly constant. At the same, however, this demonstrates that there is a certain mismatch rate remaining. The unexpected expression unit rate is about 15% when the number of learning reports is small, it drops to 6.8% when the number of learning reports increases to 500. This indicates that increasing the number of learning reports tends to reduce the number of unexpected expression units.\nConclusion:Structuring of diagnostic reports in free text format was performed using the text mining tool in order to make the contents available for secondary uses. As a result of comparison of the converted expression units with those manually generated by physicians, a match rate of 85.8% was achieved.A method for translating the contents of reports from Japanese to English using structured reports in a simple manner was also presented.EuroPACS 200