Research map / 3D vision / Face and shape
clDice - A Novel Topology-Preserving Loss Function for Tubular Structure Segmentation
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 soft-clDice to soft-Dice training | FCN and U-Net · Massachusetts Roads | clDice: 70.79 -> 76.92 (FCN); 86.83 -> 88.17 (U-Net) | Helped |
| Add soft-clDice to soft-Dice training | U-Net · CREMI | clDice: 95.86 -> 96.21 | Helped |
| Add soft-clDice to soft-Dice training | FCN and U-Net · DRIVE retina | clDice: 78.02 -> 80.95 (FCN); 75.71 -> 76.86 (U-Net) | Helped |
| Add soft-clDice to soft-Dice training | FCN and U-Net · Vessap | clDice: 90.88 -> 91.32; 90.10 -> 91.65; 91.18 -> 93.03; 90.16 -> 93.25 | Helped |
| Add soft-clDice to soft-Dice training | segmentation network · 3D synthetic vessels | clDice: 97.07 -> 98.13 | Helped |
| Add soft-clDice to BCE | segmentation network · 3D synthetic vessels | clDice: 98.24 -> 98.46 | Helped |
| Add soft-clDice to MSE | segmentation network · 3D synthetic vessels | clDice: 97.03 -> 98.37 | Helped |
| Increase the soft-clDice weight α (ablation) | soft-clDice combined loss · Massachusetts Roads | clDice: 70.79; 74.80; 76.25; 74.86; 76.92; 75.22 (FCN) | Mixed |
| Vary α from 0.1 to 0.5 (ablation) | soft-clDice combined loss · CREMI | clDice: 96.05; 96.01; 96.21; 96.09; 96.16 | No clear effect |
| Choose α per dataset (ablation) | soft-clDice combined loss · all evaluated datasets | clDice | Mixed |
| Use soft-clDice rather than soft-Dice alone | segmentation network · all evaluated datasets | topological, volumetric, and graph similarity | Helped |
| Use foreground-only clDice for highly imbalanced data rather than computing it on foreground and background (ablation) | soft-clDice · highly imbalanced tubular datasets | segmentation performance | No clear effect |
Closest papers by meaning
- GraphMorph: Tubular Structure Extraction by Morphing Predicted Graphs (NeurIPS 2024 · similarity 0.88)
- Topograph: An Efficient Graph-Based Framework for Strictly Topology Preserving Image Segmentation (ICLR 2025 · similarity 0.86)
- Efficient Connectivity-Preserving Instance Segmentation with Supervoxel-Based Loss Function (AAAI 2025 · similarity 0.84)
- Complex Wavelet Mutual Information Loss: A Multi-Scale Loss Function for Semantic Segmentation (ICML 2025 · similarity 0.84)
- DeNVeR: Deformable Neural Vessel Representations for Unsupervised Video Vessel Segmentation (CVPR 2025 · similarity 0.83)
- SegGraph: Leveraging Graphs of SAM Segments for Few-Shot 3D Part Segmentation (NeurIPS 2025 · similarity 0.82)
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.