New reporting system for the upper gastrointestinal endoscopy using speech recognition technology

Backgrounds
We have developed a new speech recognition system for an endoscopy report with FUJIFILM Medical IT solutions and NTT Corporation. This system automatically extracts only endoscopic terminology from endoscopists’ oral statements. It has not been demonstrated whether this system is practical even during the endoscopic examination in which there are various noises.
Aim
The purpose of this study is to evaluate our speech recognition system in actual clinical practice.
Methods
We input endoscopic findings into the system during each endoscopic procedure, using the speech recognition system. Three columns for an organ name, a disease name and a classification of the disease were called a row of columns in the report. The “recognition rate” was defined as the number of rows with correct findings divided by the total number of rows. The time from when the system recognized the spoken voice until the findings were registered was defined as the “registration time”.
Results
One hundred twenty-seven findings were registered from 34 subjects who obtained their consents for this study. The recognition rate was 100%. The number of corrections of the findings was 0. All the findings were correctly recognized and registered once. The median registration time was 0.024 seconds.
Conclusion
Our speech recognition system had high accuracy and fast response time even in a noisy clinical setting.