[Generation of brain scintigram reports with a speech recognition device controlled by a personal computer]

Radiology reporting system using a speech recognition apparatus was applied to generation of reports on brain scintigram and SPECT. The number of words or phrases that were selected for the reporting system by the authors was 259. Rates of correct recognition at the first, second, and third or more utterance of the voices were 91.1, 5.0, and 3.9%, respectively. The recognition rates decreased from 96.6 to 88.0% when the mean lengths of the words or phrases varied from 6.4 to 3.3. Net time for the generation of 8 reports by the system was 21 minutes 26 seconds, while that by the traditional dictation-transcription method was 15 minutes and 53 seconds. This system seems to be useful because of much longer input time expected for an average doctor, who usually is not well trained in using a Japanese-word processor, and convenience for database management offered by the system.