Correspondences between microcalcification projections on two mammographic views acquired with digital systems

Area of Science:

• Medical Imaging
• Computer Vision
• Biomedical Engineering

Background:

• Accurate matching of microcalcifications (MC) across mammogram pairs is crucial for 3D reconstruction.
• Existing methods may struggle with breast deformation, affecting the precision of MC localization.

Purpose of the Study:

• To develop and evaluate an automated algorithm for matching MC projections on two mammograms of the same breast.
• To assess the algorithm's performance in comparison to manual matching and its utility in 3D MC cluster reconstruction.

Main Methods:

• A three-step algorithm utilizing morphological features and similarity functions to match MCs.
• Incorporation of geometrical constraints and an epipolar strip to account for breast deformation.
• Comparison of automated matching results against expert manual matching and reconstructed cluster shapes.

Main Results:

• The automated algorithm achieved a 78.43% concordance rate with confident manual MC matching.
• Reconstruction of MC clusters using the automated algorithm showed variable resemblance to manually reconstructed clusters, indicating areas for improvement.

Conclusions:

• The proposed algorithm demonstrates potential for automated MC matching in mammography, with promising results for 3D reconstruction applications.
• Further refinement, potentially by incorporating additional views, is recommended to enhance accuracy and reliability.