Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

RDTE-UNet: A Boundary and Detail Aware UNet for Precise Medical Image Segmentation

Jierui Qu, Jianchun Zhao

TL;DR

RDTE-UNet targets boundary ambiguity and fine-structure preservation in medical image segmentation by unifying local feature extraction with global context. It introduces a ResBlock–Details Transformer backbone augmented by three modules: Adaptive Shape-aware Boundary Enhancement (ASBE), Horizontal–Vertical Detail Attention (HVDA) with StairConv, and Euler Feature Fusion (EulerFF) to enable multi-scale, anisotropic boundary and detail modeling. The approach achieves state-of-the-art or competitive results on Synapse and BUSI, with ablation studies confirming that each module—ASBE, HVDA, and EulerFF—contributes to improved segmentation accuracy and boundary fidelity. This work advances boundary-aware segmentation with a cohesive architecture that can better handle morphological variability and complex topologies, potentially improving clinical decision support.

Abstract

Medical image segmentation is essential for computer-assisted diagnosis and treatment planning, yet substantial anatomical variability and boundary ambiguity hinder reliable delineation of fine structures. We propose RDTE-UNet, a segmentation network that unifies local modeling with global context to strengthen boundary delineation and detail preservation. RDTE-UNet employs a hybrid ResBlock detail-aware Transformer backbone and three modules: ASBE for adaptive boundary enhancement, HVDA for fine-grained feature modeling, and EulerFF for fusion weighting guided by Euler's formula. Together, these components improve structural consistency and boundary accuracy across morphology, orientation, and scale. On Synapse and BUSI dataset, RDTE-UNet has achieved a comparable level in terms of segmentation accuracy and boundary quality.

RDTE-UNet: A Boundary and Detail Aware UNet for Precise Medical Image Segmentation

TL;DR

RDTE-UNet targets boundary ambiguity and fine-structure preservation in medical image segmentation by unifying local feature extraction with global context. It introduces a ResBlock–Details Transformer backbone augmented by three modules: Adaptive Shape-aware Boundary Enhancement (ASBE), Horizontal–Vertical Detail Attention (HVDA) with StairConv, and Euler Feature Fusion (EulerFF) to enable multi-scale, anisotropic boundary and detail modeling. The approach achieves state-of-the-art or competitive results on Synapse and BUSI, with ablation studies confirming that each module—ASBE, HVDA, and EulerFF—contributes to improved segmentation accuracy and boundary fidelity. This work advances boundary-aware segmentation with a cohesive architecture that can better handle morphological variability and complex topologies, potentially improving clinical decision support.

Abstract

Medical image segmentation is essential for computer-assisted diagnosis and treatment planning, yet substantial anatomical variability and boundary ambiguity hinder reliable delineation of fine structures. We propose RDTE-UNet, a segmentation network that unifies local modeling with global context to strengthen boundary delineation and detail preservation. RDTE-UNet employs a hybrid ResBlock detail-aware Transformer backbone and three modules: ASBE for adaptive boundary enhancement, HVDA for fine-grained feature modeling, and EulerFF for fusion weighting guided by Euler's formula. Together, these components improve structural consistency and boundary accuracy across morphology, orientation, and scale. On Synapse and BUSI dataset, RDTE-UNet has achieved a comparable level in terms of segmentation accuracy and boundary quality.

Paper Structure

This paper contains 13 sections, 14 equations, 5 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Methods
  3. Adaptive Shape-aware Boundary Enhancement Module (ASBE)
  4. Details Transformer Block
  5. Horizontal and Vertical Details Self-Attention module (HVDA)
  6. StairConv
  7. Euler feature fusion Module (EulerFF)
  8. Experiments
  9. Datasets and Metrics
  10. Experimental results and visualization
  11. Ablation Study
  12. Conclusion
  13. ACKNOWLEDGMENTS

Figures (5)

  • Figure 1: Overview of the framework.
  • Figure 2: Schematic of HVDA Module.
  • Figure 3: Schematic of StairConv.
  • Figure 4: Schematic of Euler Module.
  • Figure 5: Qualitative comparison of different methods through visualization on Synapse dataset. Our method produces fewer false positives and better preserves fine details.