A Novel CNN Architecture with Integrated Self-Attention Mechanism for Mammogram Classification: Self-Attentive Convolutional Neural Network

This study introduces a Self-Attentive Convolutional Neural Network (SACNN), a novel CNN architecture integrated with a self-attention mechanism, tailored for enhanced mammogram classification. Unlike traditional CNNs, SACNN leverages self-attention to provide a refined feature map that emphasizes salient features critical for accurate diagnosis. We evaluated SACNN on two benchmark datasets, achieving a classification accuracy improvement of 3.5% over conventional CNNs. Our findings suggest that the integration of self-attention mechanisms can significantly improve the interpretability and performance of deep learning models in medical imaging. This paper details our approach, from pre-processing and augmentation techniques to the model's training regimen, and provides a comprehensive comparison with existing methodologies. The statistical significance of SACNN's performance uplift is discussed, reinforcing its potential for clinical applications.