Related Experiment Video
Updated: Jul 19, 2026
13:44
Detection of Architectural Distortion in Prior Mammograms via Analysis of Oriented Patterns
Published on: August 30, 2013
Convolutional neural networks for mammography mass lesion classification.
Summary
Convolutional neural networks automatically learn features for mammography mass lesions, improving classification accuracy. This deep learning approach significantly outperforms traditional methods in detecting abnormalities.
Area of Science:
- Medical Imaging
- Machine Learning
- Artificial Intelligence
Background:
- Mammography image analysis traditionally relies on handcrafted features for lesion detection.
- Learning-based approaches offer an alternative for automated feature discovery.
Purpose of the Study:
- To evaluate convolutional neural networks (CNNs) for automatic feature learning in mammography mass lesions.
- To compare CNN-derived features against traditional handcrafted features for classification.
Main Methods:
- Utilized convolutional neural networks to learn image representations from mammography data.
- Integrated learned features into a subsequent classification stage.
- Benchmarked performance against state-of-the-art handcrafted feature methods.
Main Results:
- The CNN-based feature extraction approach demonstrated superior performance.
- Achieved an increase in the area under the ROC curve from 79.9% to 86% compared to traditional methods.
- Indicated the suitability of deep learning for mammography analysis.
Conclusions:
- Convolutional neural networks provide an effective strategy for automatic feature learning in mammography.
- This deep learning approach significantly enhances the classification of mammographic lesions.
- Outperforms existing state-of-the-art feature representation techniques.