Enhanced BreastNet Architecture and Comparison with State-of-the-Art Models

This paper proposes a deep learning design Enhanced BreastNet enhances the effectiveness of detection of breast cancer whilst preserving computational efficiency. New components such as residual blocks, inception modules and attention techniques are integrated into the current designs with this model. While training deeper networks, Enhanced BreastNet uses ResNet-inspired residual blocks, extracting features at multiple scales, go through GoogLeNet-inspired inception modules and improves feature extraction through attention procedures, known as Squeeze-and-Excitation blocks. Leaky ReLU and Clipped ReLU are used as activation functions in the architecture because they are highly effective in making the models more robust and helping the gradients to flow properly. The Enhanced BreastNet has highest accuracy of 94. 23% accuracy and 0. 1755 loss. This is much better than XceptionNet that achieved only 93 out of our targets. 13% accuracy. Correlation analysis proves the credibility of the model constructed as it strengthens the features extracted as well as the classification done for breast cancer indicators and they reveal significant relationships between them. It can be mentioned that the given model offers good flexibility and, thus, can be used for various medical imaging tasks apart from the detection of breast cancer.