Voice-activated retrieval of mammography reference images

Area of Science:

• Radiology
• Medical Informatics
• Computer-Aided Diagnosis

Background:

• Mammography interpretation requires access to relevant prior cases for comparison and learning.
• Current methods for retrieving similar cases can be time-consuming and may not capture all relevant contextual information.

Purpose of the Study:

• To integrate context-sensitive, computer-based educational and decision-making aids into mammography interpretation.
• To determine the clinical utility of an automated image retrieval system to guide further development.

Main Methods:

• Assembled a database of 347 digital mammography images with coded features.
• Developed an interface linking a speech recognition radiology reporting system to the image database.
• Modified the reporting system to translate dictated findings into database search terms for retrieval.

Main Results:

• Voice dictation of mammographic cases automatically retrieved relevant reference images.
• Retrieved images were organized into sets based on findings, diagnoses, or user-defined parameters.
• Mammographers evaluated the clinical utility of the system's output, confirming its potential.

Conclusions:

• Case-specific and problem-specific image sets can be automatically generated from spoken dictation.
• Subsets of retrieved images can anticipate common clinical problems, aiding interpretation.
• This automated, context-sensitive image retrieval offers continuous educational and decision support during routine interpretation.