Neil Pravin Shah MD, Presenter: Stockholder, Nuance Communications, Inc

Krishna Juluru MD, Abstract Co-Author: Nothing to Disclose

The Apple iPAD is a tablet computing platform that combines a high-resolution screen, camera, microphone, wireless internet access, and powerful processor. Since its introduction, there has been much interest in the use of this technology for visualization of radiographic images. As the feature set of DICOM viewers available for iPAD matures, there is now a growing interest in using this platform to not only view images, but to simultaneously generate reports. The purpose of this study is to evaluate the accuracy of a freely available iPAD application, as compared to an application designed for a stand-alone workstation.

Ten radiograph reports were chosen at random during a shift the Emergency Room. The corrected and transcribed reports, which did not have patient identifying information, were considered the gold standard in this study. A single radiologist then re-recorded the reports into an voice recorder. These recordings were then played into an iPAD (enabled with Dragon Dictation software) and into a traditional workstation microphone (enabled with Powerscribe software). The use of recordings ensured that there was no variability in speech presented to the two platforms. Transcribed text from the two platforms was then compared to the gold standard corrected reports for accuracy.

A generous criteria for accuracy were adopted in the evaluation of the reports—as long as the original words appeared in the transcribed report in order, they were counted as accurate. The ten reports utilizing the traditional workflow demonstrated an average accuracy of 97% (standard deviation of 2.6%) versus the mobile workflow demonstrating an accuracy of 61% (standard deviation 12.6%). The results are significantly different (t-test value < 0.005). The reports from the traditional workflow required minimal corrections, while the mobile workflow required extensive corrections.

Freely-available speech recognition software for the iPAD performs with a lower degree of accuracy than software customized for medical use on a workstation. The speech recognition performance on this platform utilizing medically-customized, commercial software needs further investigation.

Freely-available speech recognition software for the iPAD may be inadequate for use at this time in medical dictation.

Shah, N, Juluru, K, Evaluation of Speech Recognition Accuracy Using Freely Available Software for iPAD.  Radiological Society of North America 2011 Scientific Assembly and Annual Meeting, November 26 - December 2, 2011 ,Chicago IL. http://archive.rsna.org/2011/11010367.html