Speech-controlled generation of radiology reports.

Abstract

Voice entry has been successfully employed to generate radiology reports with a word recognizer with a 1,000-word lexicon capacity. About 50% of reports were able to be dictated with a single 900-word lexicon. This was split into five sections by anatomic or subspecialty application. Each was augmented to 900 words. By switching from one lexicon to another, it was possible to dictate more than 70% of reports. With exclusive use of three lexicons in subspecialty areas (gastrointestinal radiology, neuroradiology, and mammography), and with further modification of the respective vocabulary, it has been possible to employ the system 88% of the time. Twelve percent of cases included wording that was beyond the scope of the lexicon. Computer subsets that allow different translations of some words when used in different contexts have been used. Some of these are used as triggers that will print whole lines, sentences, or even complete reports. Dictation times with voice entry take about 20% longer. Recognition reliability has been greater than 95%.