Breast Cancer Detection Using Mammography: Image Processing to Deep Learning

This research presents a comprehensive deep learning–driven framework for automated breast cancer detection using mammography images, integrating advanced image preprocessing, deep feature extraction, and a meta-learning–based classification strategy. The proposed system begins with robust preprocessing that enhances mammographic clarity through grayscale conversion, contrast normalization using CLAHE, and standardized resizing to ensure consistent model input. These steps significantly improve visibility of subtle breast lesions and microcalcifications, which are critical indicators of malignancy. Following preprocessing, deep features are automatically extracted using a pre-trained ResNet-50 model, leveraging the discriminative power of the global average pooling layer to capture high-level structural and textural attributes inherent in mammographic patterns. To improve diagnostic accuracy, the framework incorporates multiple base classifiers—Decision Tree, Naive Bayes, k-Nearest Neighbors, and Support Vector Machine—each contributing diverse decision boundaries. The predicted score vectors from these classifiers are combined to form a comprehensive meta-feature set. A powerful ensemble meta-learner, implemented using XGBoost to emulate XGBoost behavior, is then trained on these fused features, enabling improved generalization and robust classification of normal and abnormal breast tissue. Performance evaluation using confusion matrix–based metrics demonstrates high accuracy, precision, sensitivity, specificity, and F1-score, confirming the system’s reliability for clinical decision support. Finally, the proposed model provides a real-time prediction for any input mammography image selected by the user, offering an interpretable and efficient diagnostic pipeline. Overall, this integrated framework delivers a strong and scalable solution for early breast cancer detection, combining classical machine learning strengths with modern deep learning capabilities for enhanced clinical applicability.

Keywords: Automated breast cancer detection, Mammography preprocessing, Deep feature extraction, Meta-learning classification, Ensemble diagnostic framework.