False-positive reduction in computer-aided mass detection using mammographic texture analysis and classification.

Area of Science:

• Medical Imaging
• Computer-Aided Diagnosis
• Machine Learning

Background:

• Computer-aided detection (CAD) systems for mammograms often generate false positives due to normal tissues mimicking masses.
• Reducing these false positives is crucial for improving the efficiency and accuracy of breast cancer diagnosis.

Purpose of the Study:

• To propose a novel segmentation-free framework to reduce false positive masses detected by CAD systems.
• To enhance feature extraction and classification for improved discrimination between normal tissues and masses in mammograms.

Main Methods:

• Investigated novel features: Hilbert's image representation, Kolmogorov-Smirnov distance, and maximum subregion descriptors.
• Implemented a feature selection step to identify the most discriminative features.
• Evaluated Random Forest, Support Vector Machine, and Decision Tree classifiers on a large dataset (10168 ROIs).

Main Results:

• The combination of proposed features significantly improved results compared to individual feature sets.
• Feature selection statistically increased accuracy for all tested classifiers.
• Random Forest achieved the highest accuracy (81.09%), closely followed by SVM (80.01%) and Decision Tree (79.12%).

Conclusions:

• The developed gray level texture features show promising accuracy for discriminating normal from abnormal regions in mammograms.
• Combining features and employing feature selection enhances diagnostic performance.
• These features can be integrated with multiresolution features to further improve false positive reduction in CAD systems for breast cancer detection.