Effective mammogram classification based on center symmetric-LBP features in wavelet domain using random forests.

Area of Science:

• Medical Imaging
• Computer Science
• Machine Learning

Background:

• Mammogram classification is vital for early breast cancer diagnosis.
• Radiologists require tools to analyze mammograms efficiently.
• Accurate classification aids in distinguishing benign from malignant cases.

Purpose of the Study:

• To propose an effective method for mammogram classification.
• To enhance the accuracy of breast cancer detection through automated analysis.
• To support radiologists in decision-making processes.

Main Methods:

• Utilized random forests combined with wavelet-based center-symmetric local binary patterns (WCS-LBP) for feature extraction.
• Employed multi-resolution CS-LBP texture characteristics from non-overlapping mammogram regions.
• Applied Support Vector Machine-Recursive Feature Elimination (SVM-RFE) for feature selection, followed by random forest classification.

Main Results:

• Achieved high performance metrics: 97.3% accuracy, 97.3% precision, 97.2% recall, and 97.2% F-measure.
• Demonstrated superior performance compared to other feature variants and state-of-the-art methods.
• Obtained a Matthews Correlation Coefficient (MCC) of 94.1%, indicating robust classification.

Conclusions:

• The proposed WCS-LBP and random forest method is effective for mammogram classification.
• This approach offers improved accuracy and reliability in breast cancer diagnosis.
• The method provides valuable support for radiologists in analyzing mammograms.