Breast cancer is the second cancer-related death cause among women worldwide. Mammography is the main screening tool for breast cancer detection. This study introduces a simple yet effective deep-learning approach for distinguishing malignant from benign masses in mammography images. Utilizing unsupervised clustering algorithms to clear image noise, and preprocessing images with custom filters yielded exceptional outcomes for both digital and film scan images. The proposed methodology allowed us to build a robust model that achieved an accuracy of 96.6% overcoming the base model by 3%.