Related Experiment Video
Updated: Jul 8, 2026
13:44
Detection of Architectural Distortion in Prior Mammograms via Analysis of Oriented Patterns
Published on: August 30, 2013
Analysis of mammographic microcalcifications using gray-level image structure features
A P Dhawan1, Y Chitre, C Kaiser-Bonasso
1Dept. of Electr. & Comput. Eng., Cincinnati Univ., OH.
IEEE Transactions on Medical Imaging
|January 1, 1996
Summary
Distinguishing benign from malignant microcalcifications in mammograms is challenging. This study developed image features and used a genetic algorithm (GA) with neural networks to accurately classify difficult cases, improving breast cancer detection.
Area of Science:
- Medical Imaging
- Biomedical Engineering
- Computational Pathology
Background:
- Mammography, clinical breast examination, and breast self-examination are current standards for breast cancer screening.
- Microcalcifications are key indicators of breast cancer, but differentiating benign from malignant ones is difficult.
- Accurate classification of difficult-to-diagnose microcalcifications is crucial for effective breast cancer management.
Purpose of the Study:
- To define and evaluate image structure features for classifying malignant microcalcifications.
- To develop a robust classification system for difficult-to-diagnose microcalcification cases.
- To compare the performance of different classification algorithms for microcalcification analysis.
Main Methods:
- Defined two categories of gray-level image structure features: texture-based (global and local) and region/cluster-based (size, number, distance).
- Utilized multivariate cluster analysis and a genetic algorithm (GA) to select optimal features from 191 difficult-to-diagnose cases.
- Employed backpropagation neural networks and parametric statistical classifiers, with Receiver Operating Characteristic (ROC) analysis for performance evaluation.
Main Results:
- A combined set of GA-selected features significantly improved classification accuracy.
- The neural network classifier demonstrated superior performance compared to linear and k-nearest neighbor (KNN) classifiers.
- The developed method effectively classified difficult-to-diagnose microcalcifications.
Conclusions:
- Image structure features, particularly when selected by a GA, are effective for classifying microcalcifications.
- Neural network-based classification offers a promising approach for improving the accuracy of breast cancer detection from mammograms.
- This methodology can aid in distinguishing malignant from benign microcalcifications, leading to better patient outcomes.
