Attention-Enhanced U-Net Models for Breast Cancer Image Analysis: A Comparative Study

Breast cancer is one of the leading causes of death among women globally, and there is a need for sophisticated diagnostic tools for early diagnosis and accurate assessment of the disease. Deep learning-based image analysis (ImageNet) has shown promising results, especially with convolutional neural networks (CNNs) for breast cancer image analysis. In this work, we studied what would happen if the U-Net architecture were combined with attention techniques to enhance breast cancer image segmentation and classification accuracy. In this work, we introduce an improved Attention U-Net network, capable of jointly performing the segmentation and classification tasks with a single network, effectively optimizing performance for medical ultrasound images. The performance of the proposed Attention-based U-Net will be compared with state-of-the-art models like ResNet-50, DenseNet-121, VGG-16, and U-Net, showing that the proposed system improves classification accuracy up to 100% and achieves 98.5% segmentation accuracy. Results when using attention mechanisms demonstrate their importance in filtering out extraneous features whilst obtaining an accurate diagnostic for breast cancer imaging.