A Novel Super-pixel Grid Mixing-Based Augmentation with a Feature Fusion of Convolutional Networks for Breast Ultrasound Image Segmentation

This work introduces a novel framework for breast ultrasound image segmentation that leverages novel super-pixel grid mixing-based augmentation, an advanced loss function named contextual differential loss (CDL), and a feature fusion network (FFNet). The proposed method aims to address the limitations of current segmentation techniques by enhancing data variability, improving boundary delineation, and ensuring comprehensive feature integration. All novel concepts are utilized during training on two publicly available breast ultrasound datasets (BUS and BUSI), and we test our proposed concepts on the respective datasets that are used in training. The results and ablation study show that the results of our proposed model outperform the state-of-the-art models and create a benchmark for the future.