Speech-controlled generation of radiology reports

Area of Science:

• Medical Informatics
• Radiology Technology

Background:

• Voice entry systems are increasingly used for medical documentation.
• Radiology report generation requires specialized vocabularies for accuracy.

Purpose of the Study:

• To evaluate the efficacy of voice entry systems in generating radiology reports.
• To assess the impact of lexicon size and specialization on dictation success rates.

Main Methods:

• Utilized a voice recognition system with a 1,000-word lexicon capacity.
• Segmented lexicons by anatomic or subspecialty applications (gastrointestinal radiology, neuroradiology, mammography).
• Employed computer subsets for context-specific word translations and report generation triggers.

Main Results:

• Dictating 50% of reports was possible with a single 900-word lexicon.
• Switching between lexicons enabled dictation of over 70% of reports.
• Specialized lexicons achieved 88% system usability, with 12% of cases requiring manual input.
• Recognition reliability exceeded 95%, though dictation time increased by 20%.

Conclusions:

• Voice entry systems, particularly with tailored lexicons, are effective for radiology report generation.
• Lexicon optimization significantly enhances dictation success rates in specialized radiology fields.
• Further development is needed to address vocabulary limitations and optimize dictation efficiency.