Speech Recognition System Generates Highly Accurate Endoscopic Reports in Clinical Practice.

Area of Science:

• Medical Informatics
• Gastroenterology

Background:

• Endoscopic reporting is traditionally done from memory post-procedure, potentially impacting accuracy and efficiency.
• Developing advanced systems for real-time medical documentation is crucial for improving healthcare workflows.

Purpose of the Study:

• To evaluate the performance and accuracy of a novel speech recognition (SR) system for generating endoscopic reports.
• To assess the efficiency of the SR system in a clinical setting with background noise.

Main Methods:

• A prospective observational pilot study involving 34 patients undergoing upper endoscopy.
• Comparison of SR system-generated reports against endoscopist's handwritten reports.
• Analysis of correct recognition rate, revisions, finding registration time, and endoscopy duration.

Main Results:

• The SR system achieved a 100% correct recognition rate for 128 findings across 22 disease names.
• A median of 0 revisions was required, with a median finding registration time of 2.57 seconds.
• The system demonstrated efficiency even in the presence of varied background noise during endoscopy.

Conclusions:

• The speech recognition system shows high accuracy and efficiency for endoscopic reporting in a clinical environment.
• The rapid finding registration time suggests significant potential for streamlining the documentation process in endoscopy.