Context-Aware Multi-Scale Polyp Segmentation Network

Abstract

Colonoscopy is the gold standard for detecting colorectal lesions and is critical for early screening and prevention of colorectal cancer. However, accurate polyp segmentation remains a challenging task due to the diverse morphology, varying sizes and indistinct boundaries of polyps. To address these challenges, we propose a Context-Aware Multi-Scale Polyp Segmentation Network (CAMSNet). Specifically, we design a Multi-Scale Interaction Module (MSI) to capture multi-scale information at each layer, enabling the model to generate aggregated feature representations that adapt to polyps of different sizes. To precisely locate polyps, we introduce a Spatial Attention Enhancement Module (SAE), which integrates global and local spatial features. After identifying the target regions, we employ a Context-Aware Refinement Module (CAR) to progressively refine the segmentation results, with particular emphasis on the blurred boundaries around polyps. Extensive experiments on five widely-used datasets demonstrate that the proposed model achieves superior segmentation accuracy and generalization performance compared to current state-of-the-art methods.