Title:PINN-EMFNet: PINN-based and Enhanced Multi-Scale Feature Fusion Network for Breast Ultrasound Images Segmentation

Abstract:With the rapid development of deep learning and computer vision technologies, medical image segmentation plays a crucial role in the early diagnosis of breast cancer. However, due to the characteristics of breast ultrasound images, such as low contrast, speckle noise, and the highly diverse morphology of tumors, existing segmentation methods exhibit significant limitations in terms of accuracy and robustness. To address these challenges, this study proposes a PINN-based and Enhanced Multi-Scale Feature Fusion Network. The network introduces a Hierarchical Aggregation Encoder in the backbone, which efficiently integrates and globally models multi-scale features through several structural innovations and a novel PCAM module. In the decoder section, a Multi-Scale Feature Refinement Decoder is employed, which, combined with a Multi-Scale Supervision Mechanism and a correction module, significantly improves segmentation accuracy and adaptability. Additionally, the loss function incorporating the PINN mechanism introduces physical constraints during the segmentation process, enhancing the model's ability to accurately delineate tumor boundaries. Comprehensive evaluations on two publicly available breast ultrasound datasets, BUSIS and BUSI, demonstrate that the proposed method outperforms previous segmentation approaches in terms of segmentation accuracy and robustness, particularly under conditions of complex noise and low contrast, effectively improving the accuracy and reliability of tumor segmentation. This method provides a more precise and robust solution for computer-aided diagnosis of breast ultrasound images.