Related Experiment Video
Updated: Jun 25, 2026
09:06
Whole-brain Segmentation and Change-point Analysis of Anatomical Brain MRI—Application in Premanifest Huntington's Disease
Published on: June 9, 2018
Combination strategies in multi-atlas image segmentation: application to brain MR data.
Xabier Artaechevarria1, Arrate Munoz-Barrutia, Carlos Ortiz-de-Solorzano
1Cancer Imaging Laboratory, Center for Applied Medical Research, University of Navarra, 31008 Pamplona, Spain. xabiarta@unav.es
IEEE Transactions on Medical Imaging
|February 21, 2009
Summary
Improving medical image segmentation accuracy requires better fusion methods. A novel generalized local weighting voting method adapts weights voxel-by-voxel, outperforming global methods for high-contrast structures in brain MRI segmentation.
Area of Science:
- Medical Image Analysis
- Computational Anatomy
- Image Segmentation
Background:
- Atlas-based medical image segmentation utilizes multiple atlases to enhance accuracy.
- Independent registration and transformation of each atlas yield multiple candidate segmentations.
- Fusion of these segmentations is crucial for a final, accurate result.
Purpose of the Study:
- To address limitations of global weighting in fusing segmentations.
- To propose and evaluate a generalized local weighting voting method for improved accuracy.
- To investigate the impact of contrast characteristics on fusion method performance.
Main Methods:
- Independent registration of multiple atlas images to a target image.
- Application of transformations to obtain candidate segmentations.
- Development and implementation of a generalized local weighting voting fusion strategy.
- Evaluation using digital phantoms and human brain MR images.
Main Results:
- Global weighting methods have significant limitations in segmentation fusion.
- The proposed generalized local weighting method adapts fusion weights voxel-by-voxel based on local performance estimation.
- Local combination strategies excel in segmenting high-contrast structures.
- Global methods demonstrate robustness to noise in low-contrast regions.
Conclusions:
- No single fusion method is optimal for all anatomical structures.
- The performance of fusion techniques is dependent on regional gray-level contrast.
- Selecting the best combination method tailored to specific structures is key for maximal segmentation accuracy.
