Knowledge-based radiologic image retrieval using axes of clinical relevance

Area of Science:

• Medical Informatics
• Artificial Intelligence in Medicine
• Radiology Information Systems

Background:

• Intelligent retrieval of radiographic images is crucial for clinical decision-making.
• Current systems may lack the nuanced understanding of clinical relevance required for efficient searching.
• Feature-coded radiographic images offer potential for advanced search capabilities.

Purpose of the Study:

• To describe a novel approach for computer-based intelligent retrieval of feature-coded radiographic images.
• To introduce the AXON system, a prototype knowledge-based system for chest imaging retrieval.
• To demonstrate the utility of domain knowledge in enhancing search robustness and relevance.

Main Methods:

• Partitioning the search space using clinically relevant attribute groups termed "axes of clinical relevance."
• Embedding domain knowledge as "axis heuristics" to guide the search process.
• Implementing a graded, progressive relaxation of search constraints.

Main Results:

• The AXON system effectively demonstrates the proposed approach in chest imaging.
• Searches illustrate the potential utility of "axes of clinical relevance" and "axis heuristics."
• Preliminary tests indicate promising results for the search strategies employed.

Conclusions:

• The described approach enhances the comprehensiveness and robustness of intelligent image retrieval.
• Knowledge-based systems can significantly improve the relevance ranking of retrieved images for clinicians.
• The AXON system serves as a valuable prototype for future developments in medical image retrieval.