Mammogram segmentation using multi-atlas deformable registration.

Area of Science:

• Medical Imaging
• Computer Vision
• Machine Learning

Background:

• Accurate breast region segmentation is crucial for automated analysis in mammography, including lesion detection and telemammography.
• Traditional methods struggle with variations in image quality and breast shape.
• Machine learning approaches require extensive annotated datasets, which are often unavailable.

Purpose of the Study:

• To develop an accurate breast region segmentation method overcoming limitations of traditional and machine learning techniques.
• To combine clustering with deformable image registration for robust segmentation.
• To address the challenge of limited landmarks in breast image registration.

Main Methods:

• A clustering algorithm identifies representative atlas images capturing mammogram variations.
• Model selection determines the optimal number of clusters from low-dimensional image projections.
• Deformable image registration transfers segmentation from atlas images to new mammograms.

Main Results:

• The proposed method accurately segments various mammogram types.
• Performance was evaluated on mini-MIAS and DDSM datasets using Jaccard Index and Hausdorff Distance.
• The approach demonstrated superior performance compared to three state-of-the-art algorithms.

Conclusions:

• The combined clustering and deformable registration technique offers an effective solution for breast region segmentation.
• This method enhances the accuracy of automated mammogram analysis.
• It provides a viable alternative when large annotated datasets are not available.