LESION CLASSIFICATION IN BREAST ULTRASOUND IMAGES VIA IMAGE SEGMENTATION AND FEATURE WEIGHTING NETWORK

Breast cancer is one of the most common cancers in women worldwide, and early detection and diagnosis are crucial for improving treatment outcomes. Three types of deep learning (DL) models, including convolutional neural networks (CNNs), transformer models, and hybrid models that combine CNNs and transformers, have been widely applied to the classification of ultrasound images. However, these models still face the bottleneck of extracting critical features, and limited generalization in practical applications. Therefore, we propose an image segmentation and feature weighting method for the three-class classification of breast cancer nodules. Our approach uses two modules: the segmentation module and weight regularization, which enhance global feature extraction by adjusting the weights of the segmented image features. To evaluate performance, the proposed method was compared to three classic models, including VGG, ResNet, and Vision Tansformer (ViT), on two public datasets. The experimental results demonstrated that the proposed method achieved the highest overall accuracy of 0.97, significantly outperforming the VGG (0.84), ResNet (0.85), and ViT (0.84) models. In conclusion, our proposed method can greatly assist in the early intelligent diagnosis of breast cancer in clinical settings.