Objective models of compressed breast shapes undergoing mammography

Area of Science:

• Medical Imaging
• Biomedical Engineering
• Computational Anatomy

Background:

• Accurate breast shape representation is crucial for mammography research.
• Existing models may lack the fidelity to represent diverse clinical presentations.

Purpose of the Study:

• To create objective, data-driven models of compressed breast shapes from mammograms.
• To enable the generation of new, clinically realistic breast shapes for research.

Main Methods:

• Utilized an automated edge detection algorithm to extract breast shapes from cranio-caudal (CC) and medio-lateral oblique (MLO) mammograms.
• Applied Principal Component Analysis (PCA) to identify key shape variations and develop predictive models.
• Assessed model accuracy using Average Distance Error (ADE) on an independent dataset.

Main Results:

• Developed PCA models capturing 99.2% (CC) and 98.0% (MLO) of shape variance using six principal components.
• Achieved high fidelity in reproducing breast shapes with low mean ADE (CC: 0.90 mm, MLO: 1.43 mm).
• Successfully generated new, clinically realistic breast shapes, with model performance decreasing with fewer principal components.

Conclusions:

• PCA models effectively reproduce and generate clinically relevant breast shapes for mammography.
• These models can support research in areas like scatter correction, dosimetry, and image registration.
• A supplementary spreadsheet facilitates the application of these breast shape models.