Speech Recognition in it's Adolescence

Implications: Despite large advances in the utility of speech recognition in medicine, the stated benefits are not fully achieved at this time.  The development of this technology promises significant cost savings and improvements in efficiency especially in the radiology department.  The maturation of this technology will be an important symbiosis to the maturation of imaging and other workflow technologies in radiology.  The key to successful use of speech recognition is judicious implementation.

Analysis:  Speech recognition in the radiology department has seen large strides over the past decade.  Many of the early problems have been solved and the accuracy of many of the products available continues to improve.  The promise of a fully automated, self functioning dictation system in the radiology department has not been realized however. 

In the real world practice of a very busy radiology departments current products do not have the capability to deliver.   In a busy department, a typical radiologist my dictate over two hundred cases in a given day.  This physician is dictating at a high rate.  The dictation itself may contain many mistakes that can be corrected on the fly by a transcriptionist.  The elevated levels of intuition for meaning and the ability to navigate the context of a dictation are skills that no current speech recognition product possesses.  Additionally, the ability of the transcriptionist to distinguish the actual dictation itself from instructions by the radiologist within the dictation will result in nonsense when fed through a speech recognition product.

The speech recognition products function well in a low volume setting where the radiologist has the time to self correct limited syntax error.  In a high volume practice the radiologist becomes a correctionist and devotes significant amounts of time to the process of making the reports suitable for dissemination.  High volume practices are advised to retain a limited number of transcriptionists to function as correctionists.  This will allow the speech recognition software to handle the "heavy lifting" of converting most of the spoken word to text.  The correctionist will be able to correct contextual mistakes, make changes bases on imbedded radiologist instruction, and correct for syntax errors.  The correctionist will of course be able to throughput a much larger volume of text than a traditional transcriptionis will be able to.

This, of course, does not allow the radiology department to realize the full cost savings promised by vendors.  It does however allow for a realistic implementation of this technology that tempers cost savings on transcription without unduly overburdening the radiologist in an already busy practice.  

Like many technologies in radiology, speech recognition should be recognized for what it is--a great idea with a lot of promise that is now in it's adolescence.