Boundary modelling and shape analysis methods for classification of mammographic masses

Area of Science:

• Medical Imaging
• Computer-Aided Diagnosis
• Biomedical Engineering

Background:

• Accurate classification of benign and malignant breast masses is crucial for effective cancer diagnosis.
• Traditional shape analysis methods struggle with atypical mass shapes, such as circumscribed malignant tumors and spiculated benign masses.
• Exceptions in mass shapes pose challenges for computer-aided diagnosis systems relying on common shape analysis techniques.

Purpose of the Study:

• To address the challenge of classifying breast masses with atypical shapes using advanced shape analysis.
• To develop and evaluate novel shape features for improved differentiation between benign and malignant breast masses.
• To investigate the utility of local boundary details for enhancing classification accuracy.

Main Methods:

• Employed local boundary analysis by segmenting mass boundaries into concave and convex segments.
• Developed an iterative polygonal modeling procedure to compute shape features from boundary segments.
• Introduced two new features: spiculation index (SI) and fractional concavity (fcc), alongside the global feature of compactness.

Main Results:

• The combination of SI, fcc, and compactness achieved 82% accuracy (Az=0.79) in benign/malignant classification.
• Spiculation index (SI) alone demonstrated 80% accuracy (Az=0.82) in distinguishing mass types.
• All three features combined reached 91% accuracy in classifying masses as circumscribed versus spiculated based on shape.

Conclusions:

• Novel shape features, particularly SI and fcc, enhance the computer-aided classification of breast masses.
• Local boundary analysis provides valuable information for differentiating atypical mass shapes.
• The proposed methods show promise for improving the accuracy of diagnostic systems for breast mass classification.