GPF-Net: Gated Progressive Fusion Learning for Polyp Re-Identification

2025-12-24

AI Summary

Colonoscopy polyp re-identification faces significant challenges—including low-resolution features and severe detail loss for small polyps—under cross-view and cross-device imaging conditions, hindering accurate matching of the same polyp across sessions. To address this, we propose a gated progressive fusion architecture coupled with inter-layer semantic refinement, enabling, for the first time, fully connected, gate-controlled selective fusion across multiple feature levels. Our approach integrates a deep feature pyramid with a gating mechanism, multi-level feature interaction fusion, and a multimodal collaborative learning framework to substantially enhance fine-grained representation of small polyps. On standard benchmarks, our method significantly outperforms existing single-modality ReID approaches: it achieves a 12.6% absolute improvement in mAP for small-polyp identification and cross-device matching. This advancement provides a robust technical foundation for early colorectal cancer screening and computer-aided diagnosis.

Technology Category

• Computer Vision: Multi-modal Vision
• Machine Learning: Multi-instance/Multi-view Learning
• Intelligent Robots: Multimodal Perception & Sensor Fusion

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Security and Privacy: Large-scale security measurements
• Graph Algorithms and Modeling for the Web: Representation, reconstruction, and subgraph or motif discovery in Web-related graphs

Abstract

Colonoscopic Polyp Re-Identification aims to match the same polyp from a large gallery with images from different views taken using different cameras, which plays an important role in the prevention and treatment of colorectal cancer in computer-aided diagnosis. However, the coarse resolution of high-level features of a specific polyp often leads to inferior results for small objects where detailed information is important. To address this challenge, we propose a novel architecture, named Gated Progressive Fusion network, to selectively fuse features from multiple levels using gates in a fully connected way for polyp ReID. On the basis of it, a gated progressive fusion strategy is introduced to achieve layer-wise refinement of semantic information through multi-level feature interactions. Experiments on standard benchmarks show the benefits of the multimodal setting over state-of-the-art unimodal ReID models, especially when combined with the specialized multimodal fusion strategy.

Problem

• Polyp re-identification across different views and cameras
• Fusing multi-level features to improve small polyp recognition
• Enhancing semantic refinement through gated progressive fusion strategy

Innovation

• Gated Progressive Fusion network for selective feature fusion
• Multi-level feature interactions refine semantic information layer-wise
• Specialized multimodal fusion strategy outperforms unimodal ReID models