Research map / Detection and segmentation / Weakly supervised detection and segmentation
A Topological Loss Function for Deep-Learning Based Image Segmentation Using Persistent Homology
James R. Clough, Nick Byrne, İlkay Öksüz, Veronika A. Zimmer, Julia A. Schnabel, Andrew Peter King
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 topological post-processing loss | U-net denoiser · MNIST denoising (m=8) | classifier accuracy: 80.4% -> 86.0% | Improved |
| add topological post-processing loss | U-net denoiser · MNIST denoising (m=2) | classifier accuracy: 95.7% -> 96.2% | No clear effect |
| add topological post-processing loss | U-net denoiser · MNIST denoising (m=12) | classifier accuracy: 56.9% -> 63.0% | Improved |
| add topological post-processing loss | U-net denoiser · MNIST denoising | mean squared error: 0.037 -> 0.028 | No clear effect |
| add topological loss on unlabelled images (semi-supervised) | U-net · UK Biobank LV myocardium (m=60, N_l=10) | Dice: 62.0 -> 68.6 | Improved |
| add topological loss on unlabelled images (semi-supervised) | U-net · UK Biobank LV myocardium (m=60, N_l=10) | topologically correct proportion: 25.8% -> 62.0% | Improved |
| add topological loss on unlabelled images (semi-supervised) | U-net · UK Biobank LV myocardium (m=60, N_l=100) | Dice: 82.1 -> 82.6 | No clear effect |
| topological loss instead of morphological closure (ablation) | U-net · UK Biobank LV myocardium (m=60, N_l=10) | topologically correct proportion: 44.4% -> 62.0% | Improved |
| topological loss instead of CRF post-processing (ablation) | U-net · UK Biobank LV myocardium (m=60, N_l=10) | topologically correct proportion: 22.0% -> 62.0% | Improved |
| topological loss instead of bootstrapping semi-supervision (ablation) | U-net · UK Biobank LV myocardium (m=60, N_l=10) | Dice: 67.3 -> 68.6 | Improved |
| topological loss instead of bootstrapping semi-supervision (ablation) | U-net · UK Biobank LV myocardium (m=20, N_l=100) | Dice: 87.2 -> 86.8 | No clear effect |
| add topological post-processing loss | 3D U-net · Placenta 3D ultrasound (13-fold CV) | Dice: +0.024 | Improved |
Closest papers by meaning
- Topology-Preserving Deep Image Segmentation (NeurIPS 2019 · similarity 0.93)
- Topograph: An Efficient Graph-Based Framework for Strictly Topology Preserving Image Segmentation (ICLR 2025 · similarity 0.91)
- Structure-Aware Image Segmentation with Homotopy Warping (NeurIPS 2022 · similarity 0.90)
- Conformable Convolution for Topologically Constrained Learning of Complex Anatomical Structures (AAAI 2026 · similarity 0.90)
- Topology-Aware Segmentation Using Discrete Morse Theory (ICLR 2021 · similarity 0.90)
- Towards Scalable Topological Regularizers (ICLR 2025 · similarity 0.85)
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.
