Comparative Study on Local Binary Patterns for Mammographic Density and Risk Scoring.

Area of Science:

• Medical Imaging
• Computer Vision
• Biomedical Engineering

Background:

• Breast density is a significant risk factor for breast cancer.
• High breast density can reduce mammogram accuracy for detecting abnormalities.
• Texture analysis is crucial for understanding breast tissue characteristics.

Purpose of the Study:

• To evaluate various Local Binary Pattern (LBP) variants for breast tissue classification.
• To compare the effectiveness of different LBP descriptors in mammographic density classification.
• To determine the optimal region and parameters for breast density classification using texture analysis.

Main Methods:

• Compared classic LBP, Elliptical LBP (ELBP), Uniform ELBP, Local Directional Pattern (LDP), and Mean-ELBP.
• Evaluated alternative texture analysis techniques alongside LBP variants.
• Investigated classification performance using fibroglandular disk region versus whole breast region.
• Assessed the impact of Region-of-Interest (ROI) size/location, descriptor size, and classifier choice.
• Utilized the MIAS database with ten-run ten-fold cross-validation.

Main Results:

• Elliptical Local Binary Pattern (ELBP) and Local Directional Patterns (LDP) demonstrated superior feature extraction for mammographic tissue classification.
• Directional filters proved relevant for accurate breast density classification.
• Classification based on fibroglandular disk ROIs outperformed classification using the whole breast region.

Conclusions:

• LBP variants, particularly ELBP and LDP, are effective for mammographic breast density classification.
• Focusing on the fibroglandular disk region enhances classification accuracy.
• Texture analysis using directional features holds significant potential for improving breast cancer risk assessment and mammographic interpretation.