Research map / Detection and segmentation / CNN features
Skeleton Recall Loss for Connectivity Conserving and Resource Efficient Segmentation of Thin Tubular Structures
Yannick Kirchhoff, Maximilian R Rokuss, Saikat Roy, Balint Kovacs, Constantin Ulrich, Tassilo Wald, Maximilian Zenk, Philipp Vollmuth, Jens Kleesiek, Fabian Isensee, Klaus H. Maier-Hein
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add Skeleton Recall Loss to nnUNet | nnUNet · ToothFairy | clDice: 89.16 -> 92.05 | Improved |
| add Skeleton Recall Loss to nnUNet | nnUNet · DRIVE (5-fold CV) | beta0 error: 57.00 -> 38.75 | Improved |
| add Skeleton Recall Loss to nnUNet | nnUNet · TopCoW multi-class | Dice: 85.36 -> 86.59 | Improved |
| add Skeleton Recall Loss to nnUNet | nnUNet · TopCoW multi-class | beta0 error: 0.137 -> 0.151 | Hurt |
| add Skeleton Recall Loss to nnUNet | nnUNet · Cracks (5-fold CV) | beta0 error: 0.147 -> 0.148 | No clear effect |
| add Skeleton Recall Loss to HRNet | HRNet · DRIVE (HRNet) | Dice: 80.34 -> 76.08 | Hurt |
| add Skeleton Recall Loss to HRNet | HRNet · DRIVE (HRNet) | clDice: 77.83 -> 84.20 | Improved |
| replace clDice loss soft skeleton with Skeleton Recall Loss tubed skeleton | nnUNet training · Training cost | additional training time: 88% -> 8% | Improved |
| replace clDice loss with Skeleton Recall Loss | nnUNet training · Training cost | additional VRAM: 52% -> 2% | Improved |
| replace clDice Loss with Skeleton Recall Loss on 13-class data | nnUNet · TopCoW multi-class | feasibility | Improved |
| use Topo-clDice Loss instead of clDice Loss | nnUNet · Roads | beta0 error: 6.539 -> 11.00 | Hurt |
Closest papers by meaning
- HarmonySeg: Tubular Structure Segmentation with Deep-Shallow Feature Fusion and Growth-Suppression Balanced Loss (ICCV 2025 · similarity 0.89)
- clDice - A Novel Topology-Preserving Loss Function for Tubular Structure Segmentation (CVPR 2021 · similarity 0.89)
- Segmentation of Tubular Structures Using Iterative Training with Tailored Samples (ICCV 2023 · similarity 0.89)
- NETracer: A Topology-Aware Iterative Tracing Approach for Tubular Structure Extraction (ICCV 2025 · similarity 0.88)
- GraphMorph: Tubular Structure Extraction by Morphing Predicted Graphs (NeurIPS 2024 · similarity 0.87)
- Topograph: An Efficient Graph-Based Framework for Strictly Topology Preserving Image Segmentation (ICLR 2025 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
