[Continuous speech recognition system for radiological reporting: comparison with experience of dictation]

Area of Science:

• Radiology
• Medical Informatics
• Speech Recognition Technology

Background:

• Radiological reporting traditionally relies on dictation.
• Continuous Speech Recognition (CSR) offers potential for improved efficiency.
• Enrollment-less CSR systems aim to simplify user adoption.

Purpose of the Study:

• Compare recognition accuracy between experienced and inexperienced radiologists using an enrollment-less CSR system.
• Evaluate the overall usefulness of the CSR system in radiological reporting.

Main Methods:

• Twenty board-certified radiologists were divided into experienced (Group A, >6 years dictation) and inexperienced (Group B, no dictation experience) groups.
• Participants dictated radiological reports using a free-style approach with an enrollment-less CSR system (AmiVoice).
• Word recognition accuracy rates were calculated by counting errors and total words.

Main Results:

• The average word recognition accuracy was 96.42% +/- 1.68% for experienced radiologists and 95.92% +/- 1.15% for inexperienced radiologists.
• No statistically significant difference in accuracy rates was observed between the two groups.
• The system demonstrated high accuracy for both experienced and novice users.

Conclusions:

• Radiological report accuracy using enrollment-less CSR is independent of prior dictation experience.
• The enrollment-less CSR system is deemed convenient and useful for radiological reporting.
• This technology has the potential to streamline the reporting process for all radiologists.