Speech processing in radiology

The goal of this study was to evaluate the recognition rate, learning poten tial and amount of time needed to complete a report with the Philips; speec h recognition system SP 6000 (Philips, Best, The Netherlands). Four radiolo gists dictated reports of interventional radiology, MRI examinations of the musculoskeletal system and CT examinations of, the thorax and abdomen with the Philips system using the German language. The recognition rate of each report and improvement rate after each learning phase of the Philips syste m was assessed. The time needed to complete a report using the Philips syst em was then compared with the time needed to complete Ea report using the t ape-based system via a time analysis. The average recognition rate for the four radiologists using the Philips system was 79.6%, which improved to 92. 5% after the third adaptation. Initially, the average time demand to dictat e and correct one report was approximately 16.8 min, but this time decrease d to 8.1 min after the third adaptation. In contrast, only 3.6 min were nee ded to dictate and correct one report using the tape-based system. However, with the speech recognition system, dictation, correction and transcriptio n of the report can be completed within 15 min, whereas with the tape-based system, it takes nearly 1 day. With the Philips system, speech recognition can reach as high as 95% since each adaptation of the system improves the recognition rate by approximately 5%. While the Philips system is associate d with longer dictation times than the tape-based system, turn-around time for a complete report is substantially shorter with the Philips system than the tape-based system.