Brain tissue classification of magnetic resonance images using partial volume modeling

Area of Science:

• Medical Imaging
• Neuroscience
• Computer Vision

Background:

• Magnetic Resonance (MR) imaging is crucial for brain analysis.
• Partial volume effects in MR images complicate accurate tissue classification.
• Distinguishing gray matter, white matter, and cerebrospinal fluid is essential.

Purpose of the Study:

• To develop a fully automatic 3D classification algorithm for brain tissues in MR images.
• To address and overcome challenges posed by partial volume effects.
• To accurately segment and reclassify brain tissues, including mixed classes.

Main Methods:

• A statistical mixture model was developed and simulated to represent mixed tissue classes.
• The D'Agostino-Pearson normality test assessed the Gaussian approximation of the mixture model.
• A two-step classification process was employed: segmentation using the mixture model and reclassification using pure class knowledge.
• Both steps utilized Markov random field (MRF) models, enhanced with multifractal dimensions for improved discrimination.

Main Results:

• The proposed mixture model can be approximated by a Gaussian function under specific conditions.
• The algorithm successfully segmented brain images into pure and mixed tissue classes.
• Reclassification of mixed classes improved the accuracy of brain tissue segmentation.
• The inclusion of multifractal dimensions enhanced the discrimination of mixed classes within MRFs.

Conclusions:

• The developed algorithm provides a robust solution for automatic 3D brain tissue classification in MR images.
• The method effectively handles partial volume effects, a common challenge in MR image analysis.
• The integration of MRFs and multifractal dimensions offers improved accuracy and discrimination for complex tissue mixtures.
• The algorithm's performance was validated on both simulated and real T1-weighted MR images.