Related Experiment Video
Updated: May 15, 2026
11:03
High-resolution In Vivo Manual Segmentation Protocol for Human Hippocampal Subfields Using 3T Magnetic Resonance Imaging
Published on: November 10, 2015
Incorporating parameter uncertainty in Bayesian segmentation models: application to hippocampal subfield volumetry
Juan Eugenio Iglesias1, Mert Rory Sabuncu, Koen Van Leemput
1Martinos Center for Biomedical Imaging, MGH, Harvard Medical School, USA.
Medical Image Computing and Computer-Assisted Intervention : MICCAI ... International Conference on Medical Image Computing and Computer-Assisted Intervention
|January 5, 2013
Summary
This study enhances Bayesian segmentation models by incorporating parameter uncertainty using Monte Carlo sampling. This improves Alzheimer's disease classification accuracy and provides error bars for hippocampal subfield segmentation.
Area of Science:
- Medical image analysis
- Computational neuroscience
- Bayesian statistics
Background:
- Bayesian segmentation models integrate anatomical knowledge with image data.
- Current methods often estimate parameters as point estimates, neglecting uncertainty.
- Accurate uncertainty quantification is crucial for robust segmentation.
Purpose of the Study:
- To improve Bayesian segmentation by accurately incorporating parameter uncertainty.
- To apply Monte Carlo sampling for enhanced uncertainty estimation in segmentation.
- To evaluate the impact of improved uncertainty on Alzheimer's disease classification.
Main Methods:
- Implemented Monte Carlo sampling to capture uncertainty in free parameters.
- Applied the technique to atlas warp sampling in hippocampal subfield segmentation.
- Utilized the enhanced segmentation for an Alzheimer's disease classification task.
Main Results:
- Demonstrated significant improvement in Alzheimer's disease classification accuracy.
- Showcased the ability to provide informative "error bars" for segmentation results.
- Validated the effectiveness of Monte Carlo sampling for uncertainty in Bayesian segmentation.
Conclusions:
- Incorporating parameter uncertainty via Monte Carlo sampling enhances Bayesian segmentation.
- The method improves downstream tasks like disease classification.
- Provides valuable uncertainty estimates for individual segmented substructures.
