Medical image databases for CAD applications in digital mammography: design issues

Area of Science:

• Medical Imaging
• Artificial Intelligence in Healthcare
• Radiology Informatics

Background:

• Evaluating computer-assisted diagnosis (CADx) algorithms in digital mammography necessitates robust image databases for comparative analysis.
• Current literature reveals a lack of standardized guidelines for establishing these essential databases.
• Image selection for existing databases often relies on subjective judgment or simple availability, hindering objective algorithm assessment.

Purpose of the Study:

• To review and discuss critical issues surrounding the creation and utilization of digital mammography image databases for CADx algorithm evaluation.
• To propose potential solutions and best practices for developing standardized databases that facilitate reliable algorithm comparisons.
• To address the ongoing challenges related to optimal database size, image resolution, and content selection.

Main Methods:

• Literature review of existing guidelines and practices for digital mammography database creation.
• Analysis of subjective and availability-based image selection methods.
• Discussion of key database parameters: size, resolution, and content.

Main Results:

• Identified a significant lack of consensus on database development guidelines in the field.
• Highlighted the limitations of subjective image selection and the need for objective criteria.
• Recognized the importance of common, accessible databases for reproducible algorithm evaluation.

Conclusions:

• Standardized digital mammography databases are essential for the reliable clinical validation of CADx algorithms.
• Addressing database size, resolution, and content is critical for meaningful algorithm comparisons.
• Developing community-accepted database guidelines will advance the field of computer-assisted diagnosis in mammography.