A review of dynamic multi-scale feature integration and edge enhancement in medical image segmentation
Wenyang Yang, Cuijing Rong, Steven Kwok Keung Chow · The Imaging Science Journal · 2025
In medical imaging, lesions often display insufficient contrast and blurred boundaries relative to surrounding tissues. Even within the same disease category, their edge morphologies can vary markedly, posing significant challenges to achieving precise medical image segmentation. Dynamic multi-scale techniques combine adaptive mechanisms with multi-scale feature processing to improve segmentation performance. These methods adaptively adjust network parameters or structures to extract and separate targets of different scales, thereby clarifying lesion boundaries. This review examines recent advances in medical image segmentation using dynamic multi-scale U-Net architectures. Second, this paper focuses on three optimization strategies, edge enhancement, mitigation of grid artefacts and boundary blurring, and loss-function design, and we summarize the core characteristics of major medical image segmentation model variants. Finally, this paper outline potential avenues and strategies for future development. This review provides researchers with a systematic summary and reference to support innovation in medical image segmentation and promote clinical translation.