Mass segmentation of mammograms using Markov models associated with constrained clustering

Area of Science:

• Medical imaging
• Computer-aided diagnosis
• Biomedical engineering

Background:

• Accurate breast mass segmentation is crucial for early breast cancer detection and diagnosis.
• Traditional segmentation methods may struggle with complex image features and variations in breast density.

Purpose of the Study:

• To propose and evaluate novel variants of the Markov random field model for enhanced breast mass segmentation.
• To improve segmentation accuracy by incorporating constrained clustering techniques.

Main Methods:

• Development of four Markov random field model variants utilizing constrained clustering.
• Testing the proposed models on a public dataset of breast images.
• Quantitative evaluation using supervised segmentation measures and comparison with expert annotations.

Main Results:

• The proposed variants demonstrated superior visual segmentation compared to the original model.
• Lower final energy values indicated improved segmentation quality.
• Centroid initialization located approximately 90% of regions of interest, with pairwise constraints recovering up to 93% of masses.
• Segmentation quality remained consistent across different breast density levels.

Conclusions:

• Constrained clustering enhances Markov random field models for more effective breast mass segmentation.
• The proposed variants offer a robust and accurate approach for automated segmentation in mammography.
• These advancements contribute to improved computer-aided diagnosis systems for breast cancer screening.