Related Experiment Video
Updated: Jul 16, 2026
13:44
Detection of Architectural Distortion in Prior Mammograms via Analysis of Oriented Patterns
Published on: August 30, 2013
Content-based retrieval of mammograms using visual features related to breast density patterns
Sérgio Koodi Kinoshita1, Paulo Mazzoncini de Azevedo-Marques, Roberto Rodrigues Pereira
1Image Science and Medical Physics Center, Internal Medicine Department, Faculty of Medicine of Ribeirao Preto, University of Sao Paulo, avenida dos Bandeirantes, 3900, 14048-900, Ribeirão Preto, São Paulo, Brazil.
Journal of Digital Imaging
|February 24, 2007
Summary
This study developed a content-based image retrieval (CBIR) system for mammograms, achieving high precision rates. The system effectively retrieves similar mammograms based on tissue density and breast shape features.
Area of Science:
- Medical Imaging
- Computer Science
- Biomedical Engineering
Background:
- Mammography is crucial for breast cancer detection.
- Efficient retrieval of mammograms is essential for clinical support and research.
- Existing content-based image retrieval (CBIR) systems require robust feature extraction and effective classification methods.
Purpose of the Study:
- To develop and evaluate a CBIR system for mammograms.
- To implement novel feature extraction techniques for mammographic analysis.
- To assess the performance of a Kohonen self-organizing map (SOM) neural network for mammogram retrieval.
Main Methods:
- Feature extraction based on fibroglandular tissue density, breast shape, Radon domain, and granulometric measures.
- Application of shape, size, texture (histogram statistics, Haralick features, moment-based features) and novel features.
- Utilizing Kohonen self-organizing map (SOM) neural network for image retrieval.
- Performance evaluation using precision and recall curves on 1,080 mammograms (CC and MLO views).
Main Results:
- Achieved precision rates between 79% and 83% for the entire image set.
- Precision rates of 78%–83% for the top 50% retrieved images.
- Precision rates of 79%–86% for the top 25% retrieved images.
- Demonstrated the effectiveness of combined feature sets and SOM for mammogram retrieval.
Conclusions:
- The developed CBIR methodology shows significant potential for integration into mammography systems.
- The system offers a promising approach for efficient and accurate retrieval of mammographic images.
- Further development could enhance diagnostic support and research capabilities in mammography.
