LBA-Net: Lightweight Boundary-Aware Network for Robust Breast Ultrasound Image Segmentation

Breast ultrasound (BUS) image segmentation is crucial for tumor diagnosis but remains challenging due to noise, low contrast, and blurred boundaries. While deep learning has improved accuracy, most models are too heavy for real-time use in portable ultrasound devices. We propose LBA-Net, a lightweight boundary-aware network that addresses these challenges through a carefully designed architecture. To ensure high efficiency suitable for edge deployment, we adopt a MobileNetV3-Small encoder, which extracts hierarchical features with minimal computational cost. To capture the multi-scale contextual information essential for managing variable tumor sizes and ambiguous boundaries, an ASPP bottleneck is incorporated. Additionally, a novel LBA-Block is proposed, which combines efficient channel and spatial attention mechanisms to enhance feature representation by adaptively emphasizing informative regions and suppressing noise. A dual-head supervision strategy with a boundary-sensitive loss further improves edge precision and model robustness. On the BUSI dataset, LBA-Net achieves a Dice score of 81.4% (validation) with only 2.1M parameters and 4.8 GFLOPs, running at 122 FPS on GPU. These results demonstrate that LBA-Net offers an effective balance of accuracy, efficiency, and robustness, making it well-suited for real-time clinical deployment.