Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

3D Medical Imaging Segmentation on Non-Contrast CT

Canxuan Gang, Yuhan Peng

TL;DR

Non-contrast CT (NCCT) segmentation in 3D medical imaging is challenged by limited ground-truth data and the need for accurate volumetric labeling. The paper surveys CNN-based and Transformer-based segmentation methods, analyzes their strengths and weaknesses, and advocates a fully convolutional-free Transformer encoder–MLP decoder pipeline to enhance global context modeling and mask generation. It highlights nnUNet as the current benchmark SOTA, discusses thin-thick slice domain adaptation, and proposes releasing a thin-slice multi-semantic NCCT dataset alongside self-supervised and diffusion-based pretraining strategies. The work aims to advance presurgical simulation and planning by improving segmentation accuracy and robustness, and it outlines future directions for handling long-tail data and domain-specific pretraining in medical imaging.

Abstract

This technical report analyzes non-contrast CT image segmentation in computer vision. It revisits a proposed method, examines the background of non-contrast CT imaging, and highlights the significance of segmentation. The study reviews representative methods, including convolutional-based and CNN-Transformer hybrid approaches, discussing their contributions, advantages, and limitations. The nnUNet stands out as the state-of-the-art method across various segmentation tasks. The report explores the relationship between the proposed method and existing approaches, emphasizing the role of global context modeling in semantic labeling and mask generation. Future directions include addressing the long-tail problem, utilizing pre-trained models for medical imaging, and exploring self-supervised or contrastive pre-training techniques. This report offers insights into non-contrast CT image segmentation and potential advancements in the field.

3D Medical Imaging Segmentation on Non-Contrast CT

TL;DR

Non-contrast CT (NCCT) segmentation in 3D medical imaging is challenged by limited ground-truth data and the need for accurate volumetric labeling. The paper surveys CNN-based and Transformer-based segmentation methods, analyzes their strengths and weaknesses, and advocates a fully convolutional-free Transformer encoder–MLP decoder pipeline to enhance global context modeling and mask generation. It highlights nnUNet as the current benchmark SOTA, discusses thin-thick slice domain adaptation, and proposes releasing a thin-slice multi-semantic NCCT dataset alongside self-supervised and diffusion-based pretraining strategies. The work aims to advance presurgical simulation and planning by improving segmentation accuracy and robustness, and it outlines future directions for handling long-tail data and domain-specific pretraining in medical imaging.

Abstract

This technical report analyzes non-contrast CT image segmentation in computer vision. It revisits a proposed method, examines the background of non-contrast CT imaging, and highlights the significance of segmentation. The study reviews representative methods, including convolutional-based and CNN-Transformer hybrid approaches, discussing their contributions, advantages, and limitations. The nnUNet stands out as the state-of-the-art method across various segmentation tasks. The report explores the relationship between the proposed method and existing approaches, emphasizing the role of global context modeling in semantic labeling and mask generation. Future directions include addressing the long-tail problem, utilizing pre-trained models for medical imaging, and exploring self-supervised or contrastive pre-training techniques. This report offers insights into non-contrast CT image segmentation and potential advancements in the field.

Paper Structure

This paper contains 26 sections, 2 equations, 10 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Background
  3. Mechanism of NCCT
  4. Image Segmentation in Computer Vision
  5. Dataset and Benchmark
  6. 3D Medical Imaging Segmentation Methods
  7. Convolutional Methods
  8. Convolutional-Transformer Hybrid Methods
  9. 2D Reconstruction Method
  10. 3D Methods
  11. Experiments
  12. Datasets
  13. Validation Strategy
  14. Evaluation Metrices
  15. Dice Similarity Coefficient (DSC)
  16. ...and 11 more sections

Figures (10)

  • Figure 1: Self-attention architecture from Transformer 30
  • Figure 2: Pitch parameter in NCCT. In \ref{['fig:a']}: Pitch = 20/10 = 2. In \ref{['fig:b']}: Pitch = 10/10 = 1. In \ref{['fig:c']}: Pitch = 5/10 = 0.5.
  • Figure 3: Process of Prospective Reconstruction. Figure (a) shows a sinogram and figure (b) shows its axial view of reconstructed CT scan. 41
  • Figure 4: Averaging of three sequential slices. 45
  • Figure 5: Different interval strategies for averaging thin slices
  • ...and 5 more figures