Related Experiment Videos
Mammographic masses characterization based on localized texture and dataset fractal analysis using linear, neural and
Michael E Mavroforakis1, Harris V Georgiou, Nikos Dimitropoulos
1University of Athens, Informatics Department, TYPA buildings, University Campus, 15771 Athens, Greece. mmavrof@di.uoa.gr
Artificial Intelligence in Medicine
|May 24, 2006
Summary
This study introduces a quantitative approach for mammographic mass classification using advanced texture analysis and fractal dimensions. Machine learning, particularly Support Vector Machines (SVMs), achieved 83.9% accuracy, outperforming qualitative assessments.
Area of Science:
- Medical Imaging
- Computer-Aided Diagnosis
- Biomedical Engineering
Background:
- Mammographic mass characterization is crucial for breast cancer diagnosis.
- Texture analysis of mammograms is challenging but important for mass classification.
- Existing methods lack in-depth investigation due to complexity.
Purpose of the Study:
- To establish a quantitative approach for mammographic mass texture classification.
- To compare the information content of textural features with qualitative assessments by radiologists.
- To evaluate advanced classifier architectures and fractal analysis for improved diagnostic accuracy.
Main Methods:
- Applied extensive textural feature functions to 130 digitized mammograms.
- Utilized fractal analysis to compare textural datasets with qualitative descriptions.
- Employed various linear and non-linear classifiers, including SVMs, ANNs, and LDA.
- Evaluated classifier performance in classifying benign and malignant breast tumors.
Main Results:
- Textural features extracted at larger scales were more informative.
- Reduced feature subsets adequately described the feature space, comparable to expert qualitative descriptions.
- Non-linear classifiers, especially SVMs, outperformed linear classifiers.
- Optimal breast mass classification accuracy of 83.9% was achieved using SVMs based solely on textural features.
Conclusions:
- Quantitative texture analysis provides valuable information for mammographic mass classification.
- Advanced machine learning classifiers, particularly SVMs, show significant potential in this domain.
- This approach offers a robust alternative or supplement to traditional qualitative assessments by radiologists.