Voice recognition (VR) software has been gaining popularity among physicians in recent years.  Many developers of VR software claim that their products have advanced so much that they are now a viable alternative to conventional transcription services.  Because of the importance of rapid turnaround times in radiology, many radiology departments and radiology groups have tried adopting voice recognition technology as an alternative to conventional radiology transcription services in an attempt to shorten turnaround times and cut costs.  Most radiologists who have used voice recognition technology don't deny that it can be an alternative to radiology transcription, but at what cost?

A study by Pezullo and colleagues of the Department of Diagnostic Imaging at the Warren Alpert School of Medicine, Brown University, compared conventional radiology transcription services with a commercially available voice recognition system in private practice.  Researchers came up with the following results:

1. Reports dictated with voice recognition took 50% longer to dictate despite being 24% shorter than those conventionally transcribed.
2. There were 5.1 errors per case, and 90% of all voice recognition dictations contained errors prior to report sign-off while 10% of transcribed reports contained errors.
3. After sign-off, 35% of VR reports still had errors.

Researchers also noted that cost savings using voice recognition systems in non-academic settings may not be realized.  Based on average radiologist and transcriptionist salaries, the additional time spent dictating with voice recognition costs an additional $6.10 per case or $76,250.00 yearly. The opportunity costs may be higher. Informally surveyed, all radiologists expressed dissatisfaction with voice recognition with feelings of frustration, and increased fatigue.

The results of this study confirm my personal opinion on this matter.  Over the years, I have worked with tens of radiologists and edited thousands of radiology reports.  I have edited reports produced by many different voice recognition systems and reached the conclusion that, although different systems vary considerably in reliability, very few radiology reports produced through voice recognition are error-free and virtually all of them require editing at some level.  All VR systems rely on user cooperation.  The radiologist needs to put effort into modifying his/her dictation style to maximize the effectiveness of the VR system.  It is true that the amount of effort depends to some extent on the radiologist's dictation style, yet some effort is almost always needed.  Moreover, the interpretation of some radiologists' dictations requires a level of reasoning only the human mind is capable of.

In summary, the proposed benefits of voice recognition in radiology reporting do not come cheap.  They can only be realized through utilizing radiologists as transcriptionists which is, as any radiologist or transcriptionist will tell you, not a very effective use of resources.  It remains to be seen whether voice recognition software will ever be advanced enough to be the preferred alternative to radiology transcription.

Editorial Team