Radiology speech recognition: workflow, integration, and productivity issues

Continuous voice recognition is now a reality, with at least 3 major vendors providing independent systems for radiology departments. However, to be truly useful, such systems must be integrated into the overall medical informatics architecture including order entry into the radiology information system, the ability to receive orders from the radiology information system, acceptance and transcription of dictation, and transmission of validated reports to the hospital information system (HIS)for billing capture. In addition, the system should permit sufficient flexibility to permit radiologists to operate in multiple modes: real-time recognition, batch processing, or spooling of audio files for later transcription by human agents, which is especially useful for very difficult accents. This article demonstrates the various modes of operation for 2 recognition systems that have been used in our center, describes the basics of their recognition models, examines the HL7 messages required for medical informatics integration, and finally illustrates the effect on report turnaround times.