Refining breast cancer classification: Customized attention integration approaches with dense and residual networks for enhanced detection.

Objective: Breast cancer detection is critical for timely and effective treatment, and automatic detection systems can significantly reduce human error and improve diagnosis speed. This study aims to develop an accurate and robust framework for classifying breast cancer into benign and malignant categories using a novel machine learning architecture. Methods: We propose a dense-ResNet attention integration (DRAI) architecture that combines DenseNet and ResNet models with three attention mechanisms to enhance feature extraction from the BreakHis dataset. The attention mechanisms focus on regionally important features, improving classification accuracy. A triple-level ensemble (TLE) method combines the performance of multiple models, further enhancing prediction accuracy. Results: The proposed DRAI architecture with TLE achieves an accuracy of 99.58% in classifying breast cancer into benign and malignant categories, surpassing existing methodologies. This high accuracy demonstrates the effectiveness of the fusion architecture and its ability to reduce manual errors in breast cancer diagnosis. Conclusion: The DRAI architecture with TLE provides a robust, automated framework for breast cancer classification. Its exceptional accuracy lays a solid foundation for future advancements in automated diagnostics and offers a reliable method for aiding early breast cancer detection.