Abstract

The research introduces a novel breast cancer (BC) detection model that integrates convolutional neural networks (CNNs) with a circular attention mechanism to improve diagnostic accuracy in mammograms. By employing a patch-based approach, mammograms are segmented into 50×50 pixel patches, allowing the model to focus on localized abnormalities. The circular attention mechanism enhances the model's ability to prioritize diagnostically significant regions, improving sensitivity and minimizing false positives and false negatives. The model incorporates preprocessing techniques and data augmentation to improve performance. In addition to mammogram analysis, the integration of histopathological data strengthens the model’s diagnostic capabilities by analysing cellular structures at the microscopic level. The combined framework achieved approximately 97% accuracy, with strong recall values, ensuring minimal false negatives. Extensive testing on CBIS-DDSM and IDC datasets validated the model’s robustness, demonstrating consistent performance across various patient samples. Visualized attention maps provided insights into the model's decision-making, enhancing its interpretability for clinical use. The patch-based analysis, combined with circular attention mechanisms, effectively identified subtle cancerous features that traditional whole-image methods may overlook. The model’s ability to dynamically focus on cancer-relevant regions improves its adaptability to real-world clinical environments. The proposed model presents a promising solution for enhancing breast cancer screening, improving diagnostic accuracy, and aiding radiologists in early cancer detection and treatment planning.

Statistics