Information-theoretic CAD system in mammography: entropy-based indexing for computational efficiency and robust...

Area of Science:

• Medical Imaging
• Computer-Aided Diagnosis
• Machine Learning

Background:

• Knowledge-based computer-assisted detection (KB-CADe) systems use template matching for mammographic mass detection.
• Increasing database size improves performance but leads to computational burden and redundancy.
• Efficient database management is crucial for maintaining diagnostic accuracy and system speed.

Purpose of the Study:

• To investigate an entropy-based indexing scheme to optimize KB-CADe performance.
• To reduce computational cost and storage requirements without compromising diagnostic accuracy.
• To evaluate the scheme as both a search and selection mechanism for the knowledge database.

Main Methods:

• Implemented an entropy-based indexing scheme for template retrieval and database selection.
• Evaluated the scheme on two distinct mammographic datasets.
• Utilized information theoretic measures, such as mutual information, for image similarity assessment.

Main Results:

• Entropy-based indexing effectively identifies relevant templates, reducing the search space for detailed analysis.
• A selective deposit strategy, prioritizing high entropy cases, leads to a more concise and effective knowledge database.
• Computational costs were reduced by 55% to 80%.

Conclusions:

• Entropy-based indexing is a viable strategy for enhancing the efficiency of KB-CADe systems.
• The proposed method significantly reduces computational load while preserving diagnostic performance.
• Optimized database management through entropy-based selection improves system maintainability and effectiveness.