From channel‐spatial attention to state space models: A review of evolving mechanisms in tumour segmentation

Abstract Objective This review delineates the evolution of attention architectures in automated tumor segmentation across three pivotal paradigms: classic Pre‐Transformer attention, dominant Transformer‐based self‐attention, and emerging Mamba‐based state space models. Methods We synthesize their functional roles—from channel enhancement to long‐range dependency modelingand critically assess hybrid structures designed for multimodal clinical oncology scenarios. Findings/Results The review establishes attention mechanisms as a foundational pillar for the next generation of intelligent segmentation tools, highlighting their potential in feature enhancement and localization. Conclusion Interrogating current challenges and charting future trajectories, these mechanisms are poised to profoundly impact the landscape of precision medicine.