Related Experiment Video
Updated: Jul 20, 2026
13:44
Detection of Architectural Distortion in Prior Mammograms via Analysis of Oriented Patterns
Published on: August 30, 2013
Classification of breast masses in mammograms using genetic programming and feature selection
R J Nandi1, A K Nandi, R M Rangayyan
1Department of Electrical Engineering and Electronics, The University of Liverpool, Brownlow Hill, Liverpool, L69 3GJ, UK.
Medical & Biological Engineering & Computing
|August 29, 2006
Summary
Genetic programming (GP) accurately classifies breast masses using mammography images. This computer-aided diagnosis approach achieved over 98% accuracy, identifying fractional concavity as a key feature for distinguishing benign from malignant tumors.
Area of Science:
- Medical Imaging
- Computational Biology
- Machine Learning
Background:
- Mammography is crucial for early breast cancer detection.
- Accurate classification of breast masses (benign vs. malignant) is vital for computer-aided diagnosis.
- Previous studies have utilized a 57-image dataset with 22 computed features.
Purpose of the Study:
- To adapt and apply genetic programming (GP) for breast mass classification.
- To implicitly perform feature selection using GP.
- To refine feature selection with statistical measures.
Main Methods:
- Utilized a dataset of 57 breast mass images with 22 features (edge-sharpness, shape, texture).
- Employed genetic programming (GP) as the primary classification technique.
- Incorporated Student's t test, Kolmogorov-Smirnov test, and Kullback-Leibler divergence for feature refinement.
Main Results:
- Achieved high classification accuracies: >99.5% for training and >98% for testing.
- Leave-one-out experiments demonstrated 97.3% success for benign and 95.0% for malignant masses.
- Fractional concavity was identified as the most significant feature, automatically selected by GP.
Conclusions:
- Genetic programming is an effective method for computer-aided breast mass classification.
- GP implicitly performs feature selection, enhancing diagnostic models.
- Fractional concavity is a highly informative feature for differentiating benign and malignant breast masses.