Related Experiment Video
Updated: Dec 21, 2025
09:06
Whole-brain Segmentation and Change-point Analysis of Anatomical Brain MRI—Application in Premanifest Huntington's Disease
Published on: June 9, 2018
12.5K
Joint inference on structural and diffusion MRI for sequence-adaptive Bayesian segmentation of thalamic nuclei with
Juan Eugenio Iglesias1,2,3, Koen Van Leemput2,4, Polina Golland3
1Centre for Medical Image Computing, Department of Medical Physics and Biomedical Engineering, University College London, United Kingdom.
Information Processing in Medical Imaging : Proceedings of the ... Conference
|May 21, 2020
Summary
This study introduces a new method for joint segmentation of structural and diffusion MRI (sMRI/dMRI) data. The novel algorithm improves the accuracy of brain structure segmentation, particularly for thalamic nuclei, by fusing multi-modal imaging data.
Area of Science:
- Neuroimaging
- Medical Image Analysis
- Computational Neuroscience
Background:
- Current neuroimaging pipelines often segment structural MRI (sMRI) and diffusion MRI (dMRI) independently.
- This independent approach can limit the accuracy of segmenting certain brain structures, such as thalamic nuclei.
- Fusing sMRI and dMRI data offers potential for more precise segmentation of complex brain regions.
Purpose of the Study:
- To develop a novel algorithm for joint segmentation of multi-modal sMRI/dMRI data.
- To improve the accuracy of brain structure segmentation by leveraging the complementary information from sMRI and dMRI.
- To enable robust segmentation across various MRI acquisition parameters and contrasts.
Main Methods:
- A Bayesian segmentation framework incorporating probabilistic atlases and unsupervised appearance modeling was employed.
- A novel hierarchical likelihood term for dMRI data, utilizing Beta and Dimroth-Scheidegger-Watson distributions, was proposed.
- This dMRI likelihood was integrated with a Gaussian mixture model for sMRI data, creating a joint unsupervised likelihood.
- An inference algorithm was developed for maximum a posteriori (MAP) parameter estimation and segmentation.
Main Results:
- The proposed method was applied to segment thalamic nuclei using a histology-derived atlas on both HCP and ADNI datasets.
- The joint segmentation approach demonstrated improved accuracy compared to sMRI-only Bayesian segmentation.
- The method showed wide applicability, handling multi-modal scans with varying MRI contrasts, b-values, and directions.
Conclusions:
- Joint segmentation of sMRI and dMRI data significantly enhances the accuracy of brain structure segmentation, especially for challenging regions like thalamic nuclei.
- The developed algorithm offers a versatile and robust solution for multi-modal neuroimaging analysis.
- This approach holds promise for advancing our understanding of brain structure and function through more precise segmentation techniques.
