Enhancing Polyp Segmentation via Encoder Attention and Dynamic Kernel Update

Abstract

Polyp segmentation is a critical step in colorectal cancer detection, yet it remains challenging due to the diverse shapes, sizes, and low contrast boundaries of polyps in medical imaging. In this work, we propose a novel framework that improves segmentation accuracy and efficiency by integrating a Dynamic Kernel (DK) mechanism with a global Encoder Attention module. The DK mechanism, initialized by a global context vector from the EA module, iteratively refines segmentation predictions across decoding stages, enabling the model to focus on and accurately delineate complex polyp boundaries. The EA module enhances the network's ability to capture critical lesion features by aggregating multi scale information from all encoder layers. In addition, we employ Unified Channel Adaptation (UCA) in the decoder to standardize feature dimensions across stages, ensuring consistent and computationally efficient information fusion. Our approach extends the lesion-aware kernel framework by introducing a more flexible, attention driven kernel initialization and a unified decoder design. Extensive experiments on the KvasirSEG and CVC ClinicDB benchmark datasets demonstrate that our model outperforms several state of the art segmentation methods, achieving superior Dice and Intersection over Union scores. Moreover, UCA simplifies the decoder structure, reducing computational cost without compromising accuracy. Overall, the proposed method provides a robust and adaptable solution for polyp segmentation, with promising applications in clinical and automated diagnostic systems.