SGDC: Structurally-Guided Dynamic Convolution for Medical Image Segmentation

Bo Shi, Wei-Ping Zhu, M.N.S. Swamy

Keywords: Medical Image Segmentation, Dynamic Convolution, Structure Preservation, Feature Refinement, Boundary Awareness

TL;DR: We address the structural detail loss caused by average pooling in dynamic convolutions by introducing a pooling-free, structure-guided mechanism for medical image segmentation

Abstract:

Spatially variant dynamic convolution provides a principled way of integrating spatial adap-
tivity into deep neural networks. However, mainstream designs in medical segmentation
commonly generate dynamic kernels through average pooling, which implicitly collapses
high-frequency spatial details into a coarse, spatially-compressed representation, leading
to over-smoothed predictions that degrade the fidelity of fine-grained clinical structures.
To address this limitation, we propose a novel Structure-Guided Dynamic Convolution
(SGDC) mechanism, which leverages an explicitly supervised structure-extraction branch
to guide the generation of dynamic kernels and gating signals for structure-aware feature
modulation. Specifically, the high-fidelity boundary information from this auxiliary branch
is fused with semantic features to enable spatially-precise feature modulation. This design
replaces context aggregation with pixel-wise structural guidance, effectively preventing the
information loss in average pooling. Experimental results show that SGDC achieves state-
of-the-art performance on ISIC 2016, PH2, ISIC 2018, and CoNIC datasets, delivering
superior boundary fidelity by reducing the Hausdorff Distance (HD95) by 2.05, and pro-
viding consistent IoU gains of 0.99%-1.49% over pooling-based baselines. Moreover, the
mechanism exhibits strong potential for extension to other fine-grained, structure-sensitive
vision tasks, such as small-object detection, offering a principled solution for maintaining
structural integrity in medical image analysis. To facilitate reproducibility and encourage
further research, the implementation for both our SGE and SGDC methods is publicly
available at https://github.com/solstice0621/SGDC.

Primary Subject Area: Segmentation

Secondary Subject Area: Detection and Diagnosis

Registration Requirement: Yes

Reproducibility: https://github.com/solstice0621/SGDC

Visa & Travel: Yes

Read CFP & Author Instructions: Yes

Originality Policy: Yes

Single-blind & Not Under Review Elsewhere: Yes

LLM Policy: Yes

Midl Latex Submission Checklist: Ensure no LaTeX errors during compilation., Replace NNN with your OpenReview submission ID., Includes \documentclass{midl}, \jmlryear{2026}, \jmlrworkshop, \jmlrvolume, \editors, and correct \bibliography command., Did not override options of the hyperref package., Did not use the times package., Use the correct spelling and format, avoid Unicode characters, and use LaTeX equivalents instead., Any math in the title and abstract must be enclosed within $...$., Did not override the bibliography style defined in midl.cls and did not use \begin{thebibliography} directly to insert references., Avoid using \scalebox; use \resizebox when needed., Included all necessary figures and removed *unused* files in the zip archive., Removed special formatting, visual annotations, and highlights used during rebuttal., All special characters in the paper and .bib file use LaTeX commands (e.g., \'e for é)., No separate supplementary PDF uploads., Acknowledgements, references, and appendix must start after the main content.

Submission Number: 79