Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

HWA-UNETR: Hierarchical Window Aggregate UNETR for 3D Multimodal Gastric Lesion Segmentation

Jiaming Liang, Lihuan Dai, Xiaoqi Sheng, Xiangguang Chen, Chun Yao, Guihua Tao, Qibin Leng, Hongmin Cai, Xi Zhong

TL;DR

This work addresses the scarcity and misalignment challenges in multimodal gastric cancer MRI segmentation by introducing the GCM 2025 dataset and the HWA-UNETR framework. HWA-UNETR combines the Hierarchical Window Aggregate (HWA) block for cross-modal alignment with the Tri-orientated Fusion Mamba (TFM) block for global, multi-scale context modeling, complemented by the Stratified Group Convolution (SGC) block. Experiments on GCM 2025 and BraTS 2021 demonstrate state-of-the-art Dice performance and robust cross-modal/generalization capabilities, with Dice improvements up to $1.68\%$ on average. The work provides a significant open-source resource and methodology for advancing multimodal gastric cancer lesion segmentation, with publicly available code and dataset.

Abstract

Multimodal medical image segmentation faces significant challenges in the context of gastric cancer lesion analysis. This clinical context is defined by the scarcity of independent multimodal datasets and the imperative to amalgamate inherently misaligned modalities. As a result, algorithms are constrained to train on approximate data and depend on application migration, leading to substantial resource expenditure and a potential decline in analysis accuracy. To address those challenges, we have made two major contributions: First, we publicly disseminate the GCM 2025 dataset, which serves as the first large-scale, open-source collection of gastric cancer multimodal MRI scans, featuring professionally annotated FS-T2W, CE-T1W, and ADC images from 500 patients. Second, we introduce HWA-UNETR, a novel 3D segmentation framework that employs an original HWA block with learnable window aggregation layers to establish dynamic feature correspondences between different modalities' anatomical structures, and leverages the innovative tri-orientated fusion mamba mechanism for context modeling and capturing long-range spatial dependencies. Extensive experiments on our GCM 2025 dataset and the publicly BraTS 2021 dataset validate the performance of our framework, demonstrating that the new approach surpasses existing methods by up to 1.68\% in the Dice score while maintaining solid robustness. The dataset and code are public via https://github.com/JeMing-creater/HWA-UNETR.

HWA-UNETR: Hierarchical Window Aggregate UNETR for 3D Multimodal Gastric Lesion Segmentation

TL;DR

This work addresses the scarcity and misalignment challenges in multimodal gastric cancer MRI segmentation by introducing the GCM 2025 dataset and the HWA-UNETR framework. HWA-UNETR combines the Hierarchical Window Aggregate (HWA) block for cross-modal alignment with the Tri-orientated Fusion Mamba (TFM) block for global, multi-scale context modeling, complemented by the Stratified Group Convolution (SGC) block. Experiments on GCM 2025 and BraTS 2021 demonstrate state-of-the-art Dice performance and robust cross-modal/generalization capabilities, with Dice improvements up to on average. The work provides a significant open-source resource and methodology for advancing multimodal gastric cancer lesion segmentation, with publicly available code and dataset.

Abstract

Multimodal medical image segmentation faces significant challenges in the context of gastric cancer lesion analysis. This clinical context is defined by the scarcity of independent multimodal datasets and the imperative to amalgamate inherently misaligned modalities. As a result, algorithms are constrained to train on approximate data and depend on application migration, leading to substantial resource expenditure and a potential decline in analysis accuracy. To address those challenges, we have made two major contributions: First, we publicly disseminate the GCM 2025 dataset, which serves as the first large-scale, open-source collection of gastric cancer multimodal MRI scans, featuring professionally annotated FS-T2W, CE-T1W, and ADC images from 500 patients. Second, we introduce HWA-UNETR, a novel 3D segmentation framework that employs an original HWA block with learnable window aggregation layers to establish dynamic feature correspondences between different modalities' anatomical structures, and leverages the innovative tri-orientated fusion mamba mechanism for context modeling and capturing long-range spatial dependencies. Extensive experiments on our GCM 2025 dataset and the publicly BraTS 2021 dataset validate the performance of our framework, demonstrating that the new approach surpasses existing methods by up to 1.68\% in the Dice score while maintaining solid robustness. The dataset and code are public via https://github.com/JeMing-creater/HWA-UNETR.
Paper Structure (15 sections, 2 equations, 3 figures, 3 tables)

This paper contains 15 sections, 2 equations, 3 figures, 3 tables.

Table of Contents

  1. Introduction
  2. GCM 2025: Gastric Cancer MRI Dataset
  3. Methodology
  4. U-shaped Framework Overview
  5. Hierarchical Window Aggregate Block
  6. Stratified Group Convolution Block
  7. Tri-orientated Fusion Mamba Block
  8. Experiments
  9. Assistance Dataset
  10. Implementation Details
  11. Performance
  12. Ablation Study
  13. Conclusion
  14. Acknowledgments.
  15. Disclosure of Interests.

Figures (3)

  • Figure 1: The data visualization for GCM 2025 dataset. From top to bottom, the images demonstrate the FS-T2W, CE-T1W, and ADC modalities.
  • Figure 2: Overview of HWA-UNETR.
  • Figure 3: Visual comparisons of proposed HWA-UNETR and other SOTA methods. The arrangement of the images in this figure corresponds to that in Fig. \ref{['fig: GCM']}.