Related Experiment Video
Updated: Jan 13, 2026
10:25
Brain Infarct Segmentation and Registration on MRI or CT for Lesion-symptom Mapping
Published on: September 25, 2019
49.3K
Automatic anatomical brain MRI segmentation combining label propagation and decision fusion
Rolf A Heckemann1, Joseph V Hajnal, Paul Aljabar
1Imaging Sciences Department, MRC Clinical Sciences Centre, Imperial College at Hammersmith Hospital Campus, Du Cane Road, London W12 0HS, UK.
Neuroimage
|July 25, 2006
Summary
Automated brain image segmentation using anatomical correspondence and decision fusion significantly improves accuracy. Combining multiple segmentations enhances results, offering a practical alternative to manual methods for large studies.
Area of Science:
- Neuroimaging
- Medical Image Analysis
- Computational Anatomy
Background:
- Manual segmentation of 3D MR brain images is time-consuming and requires expertise.
- Automated methods are needed for large-scale neuroimaging studies.
- Current automated methods for brain structure labeling have limited accuracy.
Purpose of the Study:
- To develop and evaluate an automated method for brain MR image segmentation using propagation and decision fusion.
- To improve the accuracy of automated brain structure labeling compared to existing methods.
Main Methods:
- Nonrigid registration using free-form deformations to establish anatomical correspondence between brain MR images.
- Testing direct and indirect label propagation techniques.
- Applying decision fusion to combine multiple segmentations from propagated labels.
Main Results:
- Individual segmentation propagation achieved an average similarity index (SI) of 0.754.
- Decision fusion of 29 segmentations increased SI to 0.836.
- Indirect propagation yielded an SI of 0.779.
- A predictive model for quality improvement based on the number of fused segmentations was developed.
Conclusions:
- Decision fusion of propagated segmentations offers a significant accuracy improvement over individual propagations.
- The developed method is practical and competitive with manual brain delineations.
- This approach addresses the limitations of manual segmentation for large neuroimaging cohorts.
