Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

SegKAN: High-Resolution Medical Image Segmentation with Long-Distance Dependencies

Shengbo Tan, Rundong Xue, Shipeng Luo, Zeyu Zhang, Xinran Wang, Lei Zhang, Daji Ergu, Zhang Yi, Yang Zhao, Ying Cai

TL;DR

SegKAN addresses the challenge of long-distance, high-resolution hepatic vessel segmentation in CT images by introducing two key innovations. The Position-Time Series Network (PTSN) converts spatial relationships among 3D patches into temporal sequences to strengthen long-range context, while Fourier-based KAN Convolution (FKAC) stabilizes embeddings to prevent gradient explosions and reduce noise. Together, these modules form a cohesive framework that, when evaluated on the MSD Hepatic Vessel dataset, achieves a Dice score of 69.02% on average and 65.90% for vessel-specific segmentation, outperforming state-of-the-art methods such as TransUNet by up to 1.78 percentage points. The results demonstrate SegKAN’s effectiveness in modeling global context and long-range dependencies for elongated structures, with potential applicability to other high-resolution medical image segmentation tasks.

Abstract

Hepatic vessels in computed tomography scans often suffer from image fragmentation and noise interference, making it difficult to maintain vessel integrity and posing significant challenges for vessel segmentation. To address this issue, we propose an innovative model: SegKAN. First, we improve the conventional embedding module by adopting a novel convolutional network structure for image embedding, which smooths out image noise and prevents issues such as gradient explosion in subsequent stages. Next, we transform the spatial relationships between Patch blocks into temporal relationships to solve the problem of capturing positional relationships between Patch blocks in traditional Vision Transformer models. We conducted experiments on a Hepatic vessel dataset, and compared to the existing state-of-the-art model, the Dice score improved by 1.78%. These results demonstrate that the proposed new structure effectively enhances the segmentation performance of high-resolution extended objects. Code will be available at https://github.com/goblin327/SegKAN

SegKAN: High-Resolution Medical Image Segmentation with Long-Distance Dependencies

TL;DR

SegKAN addresses the challenge of long-distance, high-resolution hepatic vessel segmentation in CT images by introducing two key innovations. The Position-Time Series Network (PTSN) converts spatial relationships among 3D patches into temporal sequences to strengthen long-range context, while Fourier-based KAN Convolution (FKAC) stabilizes embeddings to prevent gradient explosions and reduce noise. Together, these modules form a cohesive framework that, when evaluated on the MSD Hepatic Vessel dataset, achieves a Dice score of 69.02% on average and 65.90% for vessel-specific segmentation, outperforming state-of-the-art methods such as TransUNet by up to 1.78 percentage points. The results demonstrate SegKAN’s effectiveness in modeling global context and long-range dependencies for elongated structures, with potential applicability to other high-resolution medical image segmentation tasks.

Abstract

Hepatic vessels in computed tomography scans often suffer from image fragmentation and noise interference, making it difficult to maintain vessel integrity and posing significant challenges for vessel segmentation. To address this issue, we propose an innovative model: SegKAN. First, we improve the conventional embedding module by adopting a novel convolutional network structure for image embedding, which smooths out image noise and prevents issues such as gradient explosion in subsequent stages. Next, we transform the spatial relationships between Patch blocks into temporal relationships to solve the problem of capturing positional relationships between Patch blocks in traditional Vision Transformer models. We conducted experiments on a Hepatic vessel dataset, and compared to the existing state-of-the-art model, the Dice score improved by 1.78%. These results demonstrate that the proposed new structure effectively enhances the segmentation performance of high-resolution extended objects. Code will be available at https://github.com/goblin327/SegKAN
Paper Structure (11 sections, 8 equations, 4 figures, 4 tables)

This paper contains 11 sections, 8 equations, 4 figures, 4 tables.

Table of Contents

  1. Introduction
  2. RELATED WORKS
  3. Methods
  4. paradigm—Position-Time Series Network (PTSN)
  5. Fourier-based KAN Convolution (FKAC)
  6. Experiments
  7. Dataset
  8. Experiment Setup
  9. Experimental Results
  10. Ablation Study
  11. Conclusions

Figures (4)

  • Figure 1: Comparison of Tumor and Vessel Segmentation Results Across Different Models.
  • Figure 2: The core algorithm framework of SegKAN, illustrates the 3D image slices and encoding of 3D patches, as well as the learning of spatial relationships between 3D patches. The unique elongated structure of Hepatic blood vessels is highlighted in the top-left corner. The bottom-right image shows the impact of different split quantities on the model's segmentation results. The model performs best when the split quantity reaches 16, indicating that an appropriate number of splits can significantly improve segmentation performance in the SegKAN model.
  • Figure 3: PTSN Structure
  • Figure 4: FKAC Structure