An Effective Ensemble Machine Learning Approach to Classify Breast Cancer Based on Feature Selection and Lesion...

Area of Science:

• Medical Imaging
• Machine Learning
• Oncology

Background:

• Breast cancer is the second most common malignancy in women.
• Mammogram screening reduces mortality, but accurate diagnosis remains challenging due to difficulties in differentiating cancerous from normal tissues.

Purpose of the Study:

• To develop a computationally efficient, feature extraction-based machine learning (ML) approach for classifying breast cancer into four categories using mammography data.
• To improve the accuracy and reliability of breast cancer diagnosis.

Main Methods:

• Mammogram preprocessing and augmentation using seven techniques.
• Region of Interest (ROI) extraction via dynamic thresholding.
• Extraction of 16 geometrical features and investigation of 11 ML algorithms.
• Development of three ensemble models using stacking, with accuracy thresholds >90%, >95%, and >96%.
• Application of five feature selection methods with 14 configurations.

Main Results:

• The Random Forest Importance algorithm identified 10 optimal features (threshold 0.045).
• The best performing model achieved 98.05% test accuracy by stacking Random Forest and XGBoost classifiers (>96% accuracy threshold).
• Consistent performance was observed across K-fold cross-validation (K=3-30).

Conclusions:

• The proposed strategy integrating image processing, feature extraction, and ML demonstrates high accuracy in breast cancer classification.
• This approach offers a promising tool for enhancing diagnostic accuracy in mammography.
• The method provides a reliable and efficient solution for categorizing breast cancer.