A.RE.S.:An Interface for Automatic Reporting by Speech

Abstract

The project and the first prototype of an interface for dictating, recording and printing radiological reports is presented. The most important feature of this interface is multimodality. The radiologist may choose among speech, keyboard and mouse to generate a report. If he is busy with hands and eyes, he can dictate most of the report, when holding and analyzing the radiographs. He can type the text and select some actions from a menu or click some meaningful icons when he has his hands free. The choice of the input modality depends on the “freedom” of the radiologist and the easiness and quickness of communication. The motivation for such a project, and the study of the impact of this system on the organisation of the radiologic department in terms of possible improvements on the reporting service, are also presented. The constraints on the radiologist-computer speech communication are analyzed. The interface has been tested with the speech modality. The speech recognizer has been trained on two main dictionaries obtained by processing chest and US reports. Four users pronounced 100 chest and 20 US reports, that have been used as a speech data-test for the automatic speech recognizer. The recognition rate of the speech recognizer, on the two dictionaries, has been 98% in the best case (speaker SP4) and 93% in the worst case (speaker SP1).