Generative compressed breast shape model for digital mammography and digital breast tomosynthesis

Area of Science:

• Medical imaging
• Biomechanical modeling
• Computer-aided diagnosis

Background:

• Accurate 3D breast shape modeling is crucial for optimizing image processing and reconstruction algorithms in mammography and digital breast tomosynthesis (DBT).
• Current breast phantoms may lack realistic shapes, potentially compromising the validity of imaging algorithms.
• Realistic breast phantoms are essential for developing and validating advanced mammography and DBT techniques.

Purpose of the Study:

• To develop a statistical point distribution model of the 3D breast shape under compression.
• Utilize principal component analysis (PCA) on structured light (SL) scans of patient compressed breasts.
• Create a generative model for realistic breast shapes applicable to medical imaging research.

Main Methods:

• Acquired structured light (SL) scans from 236 patients undergoing craniocaudal-view DBT imaging.
• Processed SL scans: cleaning, merging, simplification, and gridding.
• Applied PCA to the gridded SL scans to create a 3D breast shape model.
• Correlated model parameters with patient characteristics (age, volume, thickness, etc.).

Main Results:

• The developed PCA model achieved a median mean absolute error of 0.68 mm compared to scanned breast shapes.
• Demonstrated a perfect correlation between the first PCA coefficient and compressed breast thickness.
• The model can generate new, patient-specific breast shapes with controllable thickness.

Conclusions:

• A robust generative model of 3D breast shape under compression was successfully created.
• The model shows good agreement with DBT projection images, validating its clinical relevance.
• The generative breast shape model is publicly available for research use.