[Automatic report documentation in cardiology using a speech recognition system]

Abstract

Computer systems that can convert spoken text into written text have recently become available. In one such system, the phonetics of spoken words are compared with those of 32 000 stored words, with a statistical program helping to choose the word with the highest probability of being correct. We evaluated the practicability of the IBM Voice Type system for writing medical reports using a cardiologic vocabulary. A total of 200 medical documents were generated with a mean of 301 +/- 52 words. In the mean, 12 +/- 5 words were falsely recognized in each document, resulting in a rate of correct recognition of 95.1 +/- 2.5%. It is possible to correct a falsely recognized word by choosing an alternative word from a provided list, which worked in our case in 51% (6.1 +/- 2.8 words in each document). Correction of falsely recognized words had to be done by manual input 49% of the time (5.9 +/- 2.9 words in each document). The mean time demand for word correction amounted to 57 +/- 15 s for each document, whereas correction by manual input needed more time (37 +/- 14 s) than choosing from a list of alternative words (20 +/- 4s). A requirement for use of the Voice Type system is a reduced speech rate. Dictation of our documents took on average 260 s when done with a normal speech rate, and 400 s when done at a reduced speech rate. In conclusion, automatic writing of cardiologic reports can be done easily and with a low failure rate using the IBM Voice Type system with a cardiologic vocabulary. It takes about 3 min longer to create a medical text 1 1/2 pages long which is free of mistakes by using the Voice Type system than to simply dictate the text. Time can be saved by eliminating the need to check a preliminary report. The major advantage of automated reporting is that the written report is immediately available. For each discipline, specific vocabularies should be validated.