Healthcare Breast Cancer Using Non-Linear Activated Deep Convolutional Attention Nets

Abstract

Among women worldwide, breast cancer is among the most common forms of cancer; hence, early detection and accurate diagnosis are vitally vital. While helpful, conventional diagnostic methods may lack the accuracy needed to detect minute changes in cancer characteristics. This work suggests a novel deep convolutional neural network (CNN) architecture improved with non-linear activation functions and attention mechanisms to increase feature extraction and classification performance, so tackling the problem. The proposed method detects complex, non-linear correlations in mammography images by using the power of ReLU and Leaky ReLU. Furthermore, the CNN features an attention mechanism to focus on the most crucial areas of the images, hence improving the model's ability to distinguish benign from malignant tumours. The findings clearly show a significant rise in classification accuracy with the proposed model obtaining an accuracy of 97.8%, sensitivity of 96.4%, and specificity of 98.2%. By demonstrating a consistent AUC (Area Under the Curve) of 0.99, therefore implying exceptional discrimination between the groups, cross-valuation helps to validate the model even further. Usually with an accuracy of about 92.5%, these results exceed traditional CNN models free of attention processes. Combining non-linear activation functions with attention processes in deep convolutional neural networks reveals a possible approach to enhance breast cancer diagnosis. This method greatly increases the accuracy and dependability of the diagnosis of breast cancer, therefore enhancing the possible patient outcomes.