Problem
Research questions and friction points this paper is trying to address.
Polyp Detection
Colorectal Cancer Prevention
Efficiency Improvement
A Survey on Deep Learning for Polyp Segmentation: Techniques, Challenges and Future Trends
🤖 AI Summary
Colon polyp segmentation faces two critical bottlenecks—poor clinical generalizability across imaging devices and severe scarcity of high-quality pixel-level annotations.
Method: We systematically review over 120 deep learning models and propose the first multi-dimensional taxonomy integrating architectural design, supervision paradigms (e.g., weakly supervised, semi-supervised learning), and domain adaptation strategies. We further introduce a novel direction jointly optimizing interpretability and robustness, unifying U-Net variants, vision transformers, uncertainty modeling, and domain-adaptive techniques.
Contribution/Results: State-of-the-art methods achieve a Dice score of 89.7% on CVC-ClinicDB but suffer >22% performance degradation under cross-device evaluation. Our analysis clarifies dataset biases, inconsistencies in evaluation metrics, and key clinical deployment barriers—including annotation ambiguity, hardware heterogeneity, and real-time inference constraints. This work establishes both theoretical foundations and actionable guidelines for translating polyp segmentation algorithms from controlled benchmarks to real-world clinical practice.
Innovation
Methods, ideas, or system contributions that make the work stand out.
Deep Learning
Polyp Detection and Segmentation
Unified Resource Repository
🔎 Similar Papers
No similar papers found.
