Related Experiment Video
Updated: Jul 12, 2026
13:44
Detection of Architectural Distortion in Prior Mammograms via Analysis of Oriented Patterns
Published on: August 30, 2013
Multi-scaled morphological features for the characterization of mammographic masses using statistical classification
Harris Georgiou1, Michael Mavroforakis, Nikos Dimitropoulos
1University of Athens, Informatics Department, TYPA Buildings, University Campus, 15771 Athens, Greece. xgeorgio@di.uoa.gr
Artificial Intelligence in Medicine
|August 24, 2007
Summary
This study introduces a novel signal analysis for mammographic mass boundaries, enhancing characterization through spectral and wavelet transforms. The findings reveal that discrete Fourier transform (DFT) and discrete wavelet transform (DWT) features significantly improve classification accuracy for mass diagnosis.
Area of Science:
- Medical imaging analysis
- Digital signal processing
- Machine learning in healthcare
Background:
- Mammographic mass boundary morphology is crucial for cancer diagnosis.
- Traditional shape features may not capture all relevant information.
- Advanced signal processing can enhance feature extraction for improved classification.
Purpose of the Study:
- To develop and evaluate a comprehensive signal analysis approach for mammographic mass boundary morphology.
- To identify effective shape features from spectral and multi-scale wavelet representations.
- To improve the characterization and classification of mammographic masses.
Main Methods:
- Applied signal analysis to radial distance measurements of mammographic masses.
- Utilized discrete Fourier transform (DFT) and discrete wavelet transform (DWT) for signal representation.
- Extracted seven uniresolution feature functions and multiple shape descriptors.
- Employed various classifiers including LDA, k-NN, RBF, MLP, and SVM.
- Used fractal analysis and MANOVA for feature selection and discriminative power assessment.
Main Results:
- Features from DWT components and DFT spectrum showed high discrimination value.
- Information content in DFT and DWT features relates to texture and fine-scale details.
- Neural classifiers achieved 72.3% accuracy for shape type identification.
- Support Vector Machines (SVM) achieved 91.54% accuracy for clinical diagnosis.
Conclusions:
- Spectral and wavelet-based features significantly enhance mammographic mass characterization.
- Advanced signal processing techniques improve classification accuracy for diagnosis.
- The developed methods offer a promising approach for improving mammography analysis.
