Disconnect to Connect: A Data Augmentation Method for Improving Topology Accuracy in Image Segmentation
Not on the map. Why?
Abstract (arXiv)
Accurate segmentation of thin, tubular structures (e.g., blood vessels) is challenging for deep neural networks. These networks classify individual pixels, and even minor misclassifications can break the thin connections within these structures. Existing methods for improving topology accuracy, such as topology loss functions, rely on very precise, topologically-accurate training labels, which are difficult to obtain. This is because annotating images, especially 3D images, is extremely laborious and time-consuming. Low image resolution and contrast further complicates the annotation by causing tubular structures to appear disconnected. We present CoLeTra, a data augmentation strategy that integrates to the models the prior knowledge that structures that appear broken are actually connected. This is achieved by creating images with the appearance of disconnected structures while maintaining the original labels. Our extensive experiments, involving different architectures, loss functions, and datasets, demonstrate that CoLeTra leads to segmentations topologically more accurate while often improving the Dice coefficient and Hausdorff distance. CoLeTra's hyper-parameters are intuitive to tune, and our sensitivity analysis shows that CoLeTra is robust to changes in these hyper-parameters. We also release a dataset specifically suited for image segmentation methods with a focus on topology accuracy. CoLetra's code can be found at https://github.com/jmlipman/CoLeTra.
Key facts. Cited by 1 paper in the map (data through 13 Oct 2026). Builds on 33 of its 55 references that are in the map. First posted on arXiv 7 Mar 2025.
Builds on
33 of 55 references.
References · 33
- 2017Mask R-CNNcited by 7,266
- 2019CutMix: Regularization Strategy to Train Strong Classifiers with Localizable Featurescited by 2,882
- 2016V-Net: Fully Convolutional Neural Networks for Volumetric Medical Image Segmentationcited by 2,287
- 2018Attention U-Net: Learning Where to Look for the Pancreascited by 1,429
- 2016Context Encoders: Feature Learning by Inpaintingcited by 1,427
- 2017Random Erasing Data Augmentationcited by 1,238
- 2021Resolution-robust Large Mask Inpainting with Fourier Convolutionscited by 733
- 2020Image Segmentation Using Deep Learning: A Surveycited by 659
- 2020Learning from Noisy Labels with Deep Neural Networks: A Surveycited by 581
- 20183D MRI brain tumor segmentation using autoencoder regularizationcited by 325
- 2020clDice -- A Novel Topology-Preserving Loss Function for Tubular Structure Segmentationcited by 290
- 2019Topology-Preserving Deep Image Segmentationcited by 198
21 more
- 2019A Topological Loss Function for Deep-Learning based Image Segmentation using Persistent Homologycited by 134
- 2017Deep Learning with Topological Signaturescited by 113
- 2018DUNet: A deformable network for retinal vessel segmentationcited by 90
- 2023Dynamic Snake Convolution based on Topological Geometric Constraints for Tubular Structure Segmentationcited by 84
- 2019Feature Pyramid and Hierarchical Boosting Network for Pavement Crack Detectioncited by 80
- 2019A Topology Layer for Machine Learningcited by 79
- 2024Skeleton Recall Loss for Connectivity Conserving and Resource Efficient Segmentation of Thin Tubular Structurescited by 43
- 2024Centerline Boundary Dice Loss for Vascular Segmentationcited by 27
- 2021Structure-Aware Image Segmentation with Homotopy Warpingcited by 24
- 2021A persistent homology-based topological loss for CNN-based multi-class segmentation of CMRcited by 21
- 2022Learning Probabilistic Topological Representations Using Discrete Morse Theorycited by 20
- 2022DTU-Net: Learning Topological Similarity for Curvilinear Structure Segmentationcited by 12
- 2021Incorporating Causal Graphical Prior Knowledge into Predictive Modeling via Simple Data Augmentationcited by 10
- 2021Persistent Homology with Improved Locality Information for more Effective Delineationcited by 10
- 2024Enhancing Boundary Segmentation for Topological Accuracy with Skeleton-based Methodscited by 9
- 2021Region-wise Loss for Biomedical Image Segmentationcited by 7
- 2020Deep Small Bowel Segmentation with Cylindrical Topological Constraintscited by 6
- 2019ANDA: A Novel Data Augmentation Technique Applied to Salient Object Detectioncited by 4
- 2020IDA: Improved Data Augmentation Applied to Salient Object Detectioncited by 4
- 2024Restoring Connectivity in Vascular Segmentation using a Learned Post-Processing Modelcited by 4
- 2023Segmentation of Tubular Structures Using Iterative Training with Tailored Samplescited by 3
Built on by
Citing papers per year: 2026: 1.
1 paper.