Enhanced Breast Cancer Classification Using Attention-Augmented CNN and Multi-View Learning on the Inbreast Dataset

Breast cancer remains a leading cause of mortality among women worldwide, emphasizing the critical need for accurate and early diagnosis. Convolutional Neural Networks (CNNs) have demonstrated remarkable performance in medical image analysis, particularly in mammographic classification tasks. Building upon prior work that employed a fine-tuned VGG-16 model combined with a Support Vector Machine (SVM) classifier on the INbreast dataset, this study proposes a novel extension to enhance both accuracy and interpretability. The proposed framework integrates Convolutional Block Attention Modules (CBAM) into the CNN architecture to enable adaptive feature refinement by focusing on salient spatial and channelwise information. Additionally, a dual-stream multi-view learning approach is introduced to leverage bilateral mammographic images, capturing cross-view contextual dependencies often overlooked in single-view analysis. To further improve classification performance, a lightweight Vision Transformer (ViT-lite) replaces the traditional SVM, facilitating effective global feature modeling through self-attention. Experimental results on the INbreast dataset demonstrate a significant improvement in classification accuracy, achieving 98.4 %, along with enhanced precision, recall, and AUC scores. The proposed model not only advances the state-of-the-art in breast cancer classification but also provides a more interpretable and scalable solution, thereby contributing to the development of reliable computer-aided diagnostic tools in clinical settings.