Related Experiment Video
Updated: Apr 26, 2026
13:44
Detection of Architectural Distortion in Prior Mammograms via Analysis of Oriented Patterns
Published on: August 30, 2013
42.6K
A new and fast image feature selection method for developing an optimal mammographic mass detection scheme
Maxine Tan1, Jiantao Pu2, Bin Zheng3
1School of Electrical and Computer Engineering, University of Oklahoma, Norman, Oklahoma 73019.
Medical Physics
|August 4, 2014
Summary
This study introduces a new sequential floating forward selection (SFFS) method for optimizing computer-aided detection (CAD) in mammography. SFFS significantly improves feature selection efficacy and classifier performance compared to genetic algorithms.
Area of Science:
- Medical Imaging
- Computer-Aided Detection (CAD)
- Machine Learning
Background:
- Feature selection is critical for developing effective computer-aided detection (CAD) schemes in medical imaging.
- Optimizing classifiers for mammographic mass detection faces challenges with large feature pools.
Purpose of the Study:
- To investigate a novel approach for enhancing image feature selection and classifier optimization in CAD schemes for mammographic masses.
- To improve the efficacy of feature selection and classifier performance in mammography CAD.
Main Methods:
- A dataset of 1600 regions of interest (ROIs) with 800 positive (malignant masses) and 800 negative cases was used.
- 271 features (shape, texture, contrast, etc.) were computed, including new texture features from dilated segments.
- Four feature selection methods (Phased Searching with NEAT, SFFS, GA, SFS) were compared to optimize an artificial neural network (ANN) classifier using tenfold cross-validation.
Main Results:
- Sequential floating forward selection (SFFS) demonstrated the highest efficacy, achieving an area under the ROC curve (AUC) of 0.864 ± 0.034.
- SFFS required only 3%-5% of the computational time compared to the genetic algorithm (GA).
- Shape, local morphological, fat, and calcification-based features were most frequently selected; new texture features improved AUC for most methods.
Conclusions:
- While genetic algorithms (GA) are powerful for CAD classifier optimization, they are computationally intensive.
- The novel SFFS-based approach significantly enhances the efficacy of image feature selection for developing CAD schemes.
- SFFS offers a more efficient and effective alternative for feature selection in mammography CAD systems.
