PCASNet: Polarized Cross-scale Attention Self-attention Network for Lightweight Ultrasound Medical Image Segmentation

U-Net and its extensions have achieved significant success in medical image segmentation but face limitations in capturing global context and transferring cross-scale features. To address these challenges, we propose a lightweight network, PCASNet, which integrates the Polarized Cross-scale Attention Module (PCAS Module) and the Dynamic Multi-scale Attention Gate (DMAG). The PCAS Module enhances global context modeling by fusing features from distant spatial positions, while the DMAG improves segmentation performance by filtering redundant features and emphasizing critical information, thereby strengthening global information modeling and feature selection capabilities. Experiments on breast and thyroid ultrasound datasets demonstrate that PCASNet outperforms traditional image segmentation algorithms in both accuracy and efficiency, highlighting its potential for applications in ultrasound medical imaging.