Skip to content

AnomaliScript/scSE-nnUNet

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

110 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

scSE-nnUNet: Cervical Spine Segmentation with Attention Mechanisms

Modified version of nnUNet exploring optimal attention placement for cervical spine segmentation.

Project Evolution

Initial Hypothesis (Not Implemented)

Originally planned cervical level-aware attention with vertebra-specific pathways (C1/C2/C3-C7). Due to the failure of implementing YOLOv11 to serve as a classifier for vertebrae because of limited high-quality cervical data to train both the YOLO classifier and the origina scSE-nnUNet, this approach was not pursued.

Actual Implementation

Systematic evaluation of scSE attention block placement within nnUNet architecture:

  1. Baseline: ResEnc nnUNet (no attention) - 0.83 DSC
  2. Configuration 1: scSE in skip connections only - 0.83 DSC (no improvement)
  3. Configuration 2: scSE in decoder + bottleneck - 0.88 DSC (+5% improvement)

Key Findings

  • Skip connection attention alone did not improve segmentation performance
  • Decoder + bottleneck attention significantly improved C6/C7 boundary precision
  • Lower cervical vertebrae (C6/C7) showed greatest improvement (+4.2% DSC)
  • Demonstrates importance of attention placement for boundary refinement tasks

Technical Details

  • Based on nnU-Net v2 architecture
  • Integrated scSE (concurrent Spatial and Channel Squeeze-and-Excitation) blocks
  • Focus on cervical vertebrae (C1-C7) segmentation from CT scans

Results

Configuration Overall DSC C7 DSC
Baseline (no attention) 0.830 0.840
Skip connections 0.830 0.840
Decoder + Bottleneck 0.880 0.882

Original Work

This project is based on nnU-Net by Fabian Isensee et al.

License

This project maintains the original Apache 2.0 License from nnUNet. See LICENSE file.

Modifications and additions © 2025 Brandon Kim

Citations

If you use this work, please cite:

  • Isensee et al. - nnU-Net (original architecture)
  • Roy et al. (2018) - scSE attention mechanisms

About

Using the attention method, csSE, innovated by Roy et al., in 2018 into nnUNetv2's framework

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors

Languages