Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Learning with Geometric Priors in U-Net Variants for Polyp Segmentation

Fabian Vazquez, Jose A. Nuñez, Diego Adame, Alissen Moreno, Augustin Zhan, Huimin Li, Jinghao Yang, Haoteng Tang, Bin Fu, Pengfei Gu

TL;DR

This work tackles the challenge of capturing geometric cues in polyp segmentation by introducing a Geometric Prior-guided Module (GPM) that injects depth priors into U-Net variants. Depth maps are generated by fine-tuning Visual Geometry Grounded Transformer (VGGT) on a colonoscopy-focused ColonDepth dataset, producing domain-adapted priors that are fused into skip connections via Cross-Update and Self-Update blocks. The approach is plug-and-play, improving CNN-, Transformer-, and Mamba-based U-Net variants across five public datasets with sharper boundaries and robust performance. The method is lightweight, offering practical gains in segmentation accuracy and boundary fidelity that can benefit clinical decision support in colorectal cancer screening.

Abstract

Accurate and robust polyp segmentation is essential for early colorectal cancer detection and for computer-aided diagnosis. While convolutional neural network-, Transformer-, and Mamba-based U-Net variants have achieved strong performance, they still struggle to capture geometric and structural cues, especially in low-contrast or cluttered colonoscopy scenes. To address this challenge, we propose a novel Geometric Prior-guided Module (GPM) that injects explicit geometric priors into U-Net-based architectures for polyp segmentation. Specifically, we fine-tune the Visual Geometry Grounded Transformer (VGGT) on a simulated ColonDepth dataset to estimate depth maps of polyp images tailored to the endoscopic domain. These depth maps are then processed by GPM to encode geometric priors into the encoder's feature maps, where they are further refined using spatial and channel attention mechanisms that emphasize both local spatial and global channel information. GPM is plug-and-play and can be seamlessly integrated into diverse U-Net variants. Extensive experiments on five public polyp segmentation datasets demonstrate consistent gains over three strong baselines. Code and the generated depth maps are available at: https://github.com/fvazqu/GPM-PolypSeg

Learning with Geometric Priors in U-Net Variants for Polyp Segmentation

TL;DR

This work tackles the challenge of capturing geometric cues in polyp segmentation by introducing a Geometric Prior-guided Module (GPM) that injects depth priors into U-Net variants. Depth maps are generated by fine-tuning Visual Geometry Grounded Transformer (VGGT) on a colonoscopy-focused ColonDepth dataset, producing domain-adapted priors that are fused into skip connections via Cross-Update and Self-Update blocks. The approach is plug-and-play, improving CNN-, Transformer-, and Mamba-based U-Net variants across five public datasets with sharper boundaries and robust performance. The method is lightweight, offering practical gains in segmentation accuracy and boundary fidelity that can benefit clinical decision support in colorectal cancer screening.

Abstract

Accurate and robust polyp segmentation is essential for early colorectal cancer detection and for computer-aided diagnosis. While convolutional neural network-, Transformer-, and Mamba-based U-Net variants have achieved strong performance, they still struggle to capture geometric and structural cues, especially in low-contrast or cluttered colonoscopy scenes. To address this challenge, we propose a novel Geometric Prior-guided Module (GPM) that injects explicit geometric priors into U-Net-based architectures for polyp segmentation. Specifically, we fine-tune the Visual Geometry Grounded Transformer (VGGT) on a simulated ColonDepth dataset to estimate depth maps of polyp images tailored to the endoscopic domain. These depth maps are then processed by GPM to encode geometric priors into the encoder's feature maps, where they are further refined using spatial and channel attention mechanisms that emphasize both local spatial and global channel information. GPM is plug-and-play and can be seamlessly integrated into diverse U-Net variants. Extensive experiments on five public polyp segmentation datasets demonstrate consistent gains over three strong baselines. Code and the generated depth maps are available at: https://github.com/fvazqu/GPM-PolypSeg
Paper Structure (14 sections, 3 figures, 4 tables)

This paper contains 14 sections, 3 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Methods
  3. Depth Maps Generation
  4. Geometric Prior-guided Module (GPM)
  5. Integrating GPMs into U-Net Variants
  6. Experiments and Results
  7. Datasets
  8. Experimental Setup
  9. Experimental Results
  10. Ablation Study
  11. Computational Complexity
  12. Conclusions
  13. Compliance with ethical standards
  14. Acknowledgements

Figures (3)

  • Figure 1: The pipeline of our proposed framework. (a) Visual Geometry Grounded Transformer (VGGT) wang2025vggt is fine tuned on a simulated endoscopic image dataset, the ColonDepth dataset rau2019implicit to generate high-fidelity, domain-adapted depth maps for polyp images. (b) Structure of U-Net variants with 4 Geometric Prior-guided Modules (GPMs) to refine skip connection features. (c) The structure of the GPM module.
  • Figure 2: Visual examples of segmentations results.
  • Figure 3: Architecture of U-Net Variant w/ 4 GPMs–Top.