Breast cancer is a serious health issue, and knowledge of protein structures is very important for the early diagnosis and treatment of breast cancer. We use Convolutional Neural Networks (CNNs) in this research to analyze protein contact maps, which indicate interactions between two regions of a protein. With the help of Principal Component Analysis (PCA) for reducing the dimensionality and training a CNN-based model, we successfully distinguish proteins as cancer-related or normal. The CNN model had 86.53% accuracy on the training set and 72.98% accuracy on the test set, clearly showing its superiority in detecting cancer-associated protein structures. The high F1-score of 65.78% also shows the good balance of precision and recall. CNNs are superior to conventional machine learning models because CNNs can extract hierarchical spatial patterns automatically without hand-crafted feature engineering. In contrast to conventional approaches like Support Vector Machines (SVMs) and Random Forest (RF), which are based on pre-defined features, CNNs learn high-level representations directly from the data and are thus better suited to identify subtle structural variations in proteins. These results demonstrate the promise of CNNs for biomarker discovery, drug discovery, and computational cancer research.