DVFIP-Net: Dual View Feature Interaction Propagation Network for Polyp Segmentation

Accurate segmentation of polyps in colonoscopy examinations plays a vital role in the early diagnosis and effective treatment of colorectal cancer. However, the considerable variation in polyp sizes and shapes often results in blurred boundaries and low contrast with surrounding mucosal tissues, posing significant challenges for precise detection and segmentation. In addition, although different channels encode distinct feature representations, their intrinsic interdependencies are frequently neglected by existing methods. To address these issues we propose a novel deep learning architecture called Dual View Feature Interaction Propagation Network DVFIP-Net for highly accurate and robust polyp segmentation. Firstly,the Local–Global View Fusion (LGVF) module integrates local and global contextual information to enhance multi-scale feature representations, improving the model’s ability to capture both fine-grained polyp textures and large-scale structural variations. Secondly, the Multi-level Feature Interaction (MFI) block enhances semantic discrimination through inter-channel communication, improving differentiation between polyps and surrounding tissues. Finally, the Feature Propagation Block (FPB) aggregates multi-stage outputs to emphasize salient features and achieve precise boundary segmentation. Extensive experiments conducted on five publicly available datasets, including ETIS-LaribPolypDB, CVC300, CVC-ClinicDB, CVC-ColonDB, and Kvasir-SEG, demonstrate that DVFIP-Net consistently outperforms state-of-the-art approaches. The proposed method attains 0.942/0.896 mDice/mIoU on CVC-ClinicDB and 0.826/0.746 on CVC-ColonDB, demonstrating superior accuracy with only 19.37 GFLOPs. The proposed approach highlights superior segmentation accuracy, generalization capability, and efficiency, underscoring its potential for real-world clinical applications in colonoscopy-based polyp detection and analysis.