Related Experiment Videos
Segmentation of brain MR images through a hidden Markov random field model and the expectation-maximization algorithm
1FMRIB Centre, John Radcliffe Hospital, University of Oxford, UK. yongyue@fmrib.ox.ac.uk
IEEE Transactions on Medical Imaging
|April 11, 2001
Summary
This study introduces a novel hidden Markov random field (HMRF) model for brain MR image segmentation, improving accuracy and robustness over traditional finite mixture (FM) models by incorporating spatial information. The HMRF-EM framework offers a more reliable approach for segmenting noisy and distorted brain images.
Area of Science:
- Medical Imaging
- Computational Neuroscience
- Image Analysis
Background:
- Finite mixture (FM) models are standard for brain MR image segmentation but lack spatial information, leading to inaccuracies with noisy or distorted images.
- Artifacts like partial volume effects and bias fields compromise the reliability of traditional FM-based segmentation methods.
- Existing MRF-based methods often use MRF as a prior within FM models, limiting their full potential.
Purpose of the Study:
- To propose a novel hidden Markov random field (HMRF) model for enhanced statistical segmentation of brain MR images.
- To address the limitations of histogram-based FM models by incorporating spatial information.
- To develop a robust and accurate HMRF-EM framework for brain MR image segmentation, capable of handling image artifacts.
Main Methods:
- Developed a novel hidden Markov random field (HMRF) model, mathematically shown to be a more general form of the FM model.
- Utilized an Expectation-Maximization (EM) algorithm to fit the HMRF model, creating an HMRF-EM framework.
- Integrated bias field correction (Guillemaud and Brady, 1997) into the HMRF-EM framework for a fully automated 3D segmentation approach.
Main Results:
- The proposed HMRF-EM framework achieves accurate and robust brain MR image segmentation.
- The model effectively incorporates spatial information through the interactions of neighboring image sites.
- Demonstrated the framework's flexibility by successfully combining it with bias field correction for automated 3D segmentation.
Conclusions:
- The HMRF-EM framework provides a significant advancement over traditional FM models for brain MR image segmentation.
- This approach enhances segmentation reliability, particularly in the presence of noise and artifacts.
- The framework's modularity allows for easy integration with other image processing techniques, paving the way for fully automated analysis.