Enhancing Breast Cancer Detection with a Hybrid CNN-Transformer Deep Learning Model
Praveena Kumari M K, Vidyashree, T Namratha Padiyar · 2025
Breast cancer remains one of the leading causes of mortality among women worldwide, emphasizing the importance of early and accurate diagnosis. While Convolutional Neural Networks (CNNs) have demonstrated strong capabilities in identifying local features from mammograms, they struggle to capture global contextual information crucial for distinguishing between benign and malignant tumors. To address these limitations, we propose a hybrid deep learning model that integrates CNNs for spatial feature extraction with Transformerbased architectures employing self-attention mechanisms to capture long-range dependencies within mammographic images. The hybrid architecture enhances tumor localization and classification accuracy by combining local and global features effectively. Experimental evaluation reveals that the proposed model achieves an accuracy of 85–90%, outperforming traditional CNN-only and Transformer-only models. It also demonstrates lower false-positive and false-negative rates, which are critical for clinical decision-making. Gradient-weighted Class Activation Mapping (Grad-CAM) and attention heatmaps are utilized to improve the interpretability of predictions, making the model more transparent and trustworthy for radiologists. This model is optimized for real-time deployment through a web-based screening interface, promoting early detection and potential application in clinical workflows. Future work will explore multimodal learning by incorporating patient metadata and employing advanced Transformer variants, such as Vision Transformers (ViT) and Swin Transformers, to further enhance classification performance.