Related Experiment Video
Updated: Apr 20, 2026
06:53
3D Whole-heart Myocardial Tissue Analysis
Published on: April 12, 2017
9.5K
3D Segmentation of Multi-contrast Cardiac Magnetic Resonances With Topological Correction and Synthetic Data
Ricardo M Rosales1,2,3, Manuel Doblaré4,5,6,7, Esther Pueyo4,5,6,7
1Aragón Institute of Engineering Research (I3A), Zaragoza, Aragón, Spain. rrosales@unizar.es.
Annals of Biomedical Engineering
|April 18, 2026
Summary
Synthetic augmentation (SA) and topological correction (TC) significantly improve 3D cardiac magnetic resonance (CMR) image segmentation. Combining SA and TC enhances segmentation accuracy, especially for complex, real-world datasets, aiding clinical diagnosis.
Area of Science:
- Medical imaging analysis
- Cardiovascular imaging
- Artificial intelligence in medicine
Background:
- Automatic segmentation of cardiac magnetic resonance (CMR) images is crucial for evaluating heart structure and function.
- Generative adversarial networks (GANs) for synthetic augmentation (SA) and persistent homology for topological correction (TC) are emerging techniques to improve segmentation.
- The combined effectiveness of SA and TC for 3D CMR segmentation remains largely unexplored.
Purpose of the Study:
- To systematically evaluate the individual and combined effectiveness of SA and TC for 3D CMR segmentation.
- To assess these techniques across diverse and challenging multi-vendor, multi-center, multi-class, and multi-contrast datasets.
- To investigate improvements in segmentation accuracy and topological precision.
Main Methods:
- Utilized anisotropic, topologically inconsistent cine and late gadolinium-enhanced (LGE) CMRs, and isotropic, topologically consistent ex vivo CMRs.
- Applied topological correction (TC) by retraining a 3D convolutional neural network (CNN) with a loss function addressing topological discrepancies.
- Generated synthetic images using 3D GANs with deformed ground truth labels for synthetic augmentation (SA).
Main Results:
- Consistent segmentation improvements were observed for ex vivo data with both overlap and topological precision using SA and TC individually and combined.
- Combined SA and TC notably enhanced infarction identification in LGE data.
- SA increased prediction overlap with ground truth labels, while TC reduced topological discrepancies across all datasets.
Conclusions:
- Topological correction (TC) and synthetic augmentation (SA) show significant potential for enhancing 3D CMR segmentation.
- These methods are particularly effective on complex, real-world datasets.
- The availability of topologically consistent training data further bolsters the performance of TC and SA.
