Automated classification of parenchymal patterns in mammograms

Area of Science:

• Medical Imaging
• Radiology
• Computer-Aided Diagnosis

Background:

• Mammographic density is a significant breast cancer risk factor.
• Accurate and reproducible assessment of mammographic density is crucial for risk stratification.
• Existing methods may be sensitive to variations in mammographic imaging techniques.

Purpose of the Study:

• To develop an automated method for determining parenchymal patterns in mammograms.
• To create a technique insensitive to mammographic imaging technique variations.
• To investigate the relationship between mammographic density and breast cancer risk.

Main Methods:

• Automated segmentation of pectoral muscle using Hough transform in oblique mammograms.
• Classification of parenchymal patterns based on a distance transform subdividing breast tissue.
• Feature extraction from grey-level histograms within distance-defined regions, minimizing tissue thickness dependency.
• Utilized k-nearest neighbors (kNN) classifier with leave-one-out cross-validation on 615 digitized mammograms.

Main Results:

• Achieved 67% exact agreement with radiologist classification across four density categories.
• Only 2% of cases showed a difference of more than one category compared to radiologist assessment.
• For mammograms recorded after 1991, exact agreement reached 80%.

Conclusions:

• The developed automated method provides robust and accurate mammographic density classification.
• The technique's insensitivity to imaging variations enhances its clinical applicability.
• This automated approach has the potential to improve breast cancer risk assessment through objective density evaluation.