Related Experiment Video
Updated: Apr 30, 2026
11:03
High-resolution In Vivo Manual Segmentation Protocol for Human Hippocampal Subfields Using 3T Magnetic Resonance Imaging
Published on: November 10, 2015
8.8K
Multi-atlas segmentation of the whole hippocampus and subfields using multiple automatically generated templates
Jon Pipitone1, Min Tae M Park1, Julie Winterburn1
1Kimel Family Translational Imaging-Genetics Lab, Centre for Addiction and Mental Health, Toronto, ON, Canada.
Neuroimage
|May 3, 2014
Summary
MAGeT-Brain offers accurate automatic hippocampus segmentation using minimal atlases, outperforming multi-atlas methods. This advance in magnetic resonance imaging (MRI) analysis reduces manual effort while maintaining high segmentation quality.
Area of Science:
- Neuroimaging
- Medical Image Analysis
- Computational Neuroscience
Background:
- Multi-atlas segmentation improves magnetic resonance imaging (MRI) analysis but requires numerous manual atlases.
- Manual atlas creation is time-consuming and requires specialized expertise.
- Existing methods necessitate extensive manual segmentation efforts.
Purpose of the Study:
- To introduce MAGeT-Brain (Multiple Automatically Generated Templates), an algorithm for automatic hippocampus segmentation.
- To minimize the number of required atlases while achieving multi-atlas approach agreement.
- To enable reliable multi-atlas segmentation with laborious or specialized atlases.
Main Methods:
- MAGeT-Brain propagates atlas segmentations to a template library using nonlinear image registration.
- It fuses propagated segmentations using a label fusion method.
- Validated through Monte Carlo cross-validation experiments on Alzheimer's Disease Neuroimaging Database (ADNI) and psychosis datasets.
Main Results:
- MAGeT-Brain achieved a mean Dice's Similarity Coefficient (DSC) of 0.869 in ADNI cross-validation using only 9 atlases.
- Demonstrated significantly lower DSC variability compared to multi-atlas segmentation.
- Showed superior agreement with manual segmentation volumes compared to FreeSurfer and FSL FIRST on psychosis and ADNI datasets.
Conclusions:
- MAGeT-Brain provides consistent whole hippocampus segmentation with fewer than 10 atlases across diverse populations and MRI types.
- The algorithm effectively segments hippocampal subfields (CA1, CA4/DG, subiculum) with competitive overlap scores.
- MAGeT-Brain represents a significant advancement in efficient and accurate neuroimaging analysis.
