Causal Markov random field for brain MR image segmentation.

Area of Science:

• Medical Imaging
• Computer Vision
• Statistical Modeling

Background:

• Accurate human brain image segmentation is crucial for neurological research and clinical diagnosis.
• Existing methods like FreeSurfer provide valuable segmentation but can be improved for higher precision.
• Bayesian classifiers offer a probabilistic framework for image analysis, but novel models are needed.

Purpose of the Study:

• To introduce and evaluate a novel Bayesian classifier, MAP-QMRF, for human brain Magnetic Resonance (MR) image segmentation.
• To leverage the Quadrilateral Markov Random Field (QMRF) model for improved prior and likelihood probability estimation.
• To assess the performance of MAP-QMRF against established segmentation tools.

Main Methods:

• Developed a Maximum A Posteriori (MAP) classifier incorporating a second-order inhomogeneous anisotropic Quadrilateral Markov Random Field (QMRF).
• Modeled joint distributions using products of 2D clique distributions within the QMRF.
• Trained and validated the MAP-QMRF classifier on 20 manually labeled human brain MR images using jackknife validation.

Main Results:

• The MAP-QMRF classifier demonstrated an average gain of 1.8% in Dice overlap measure compared to FreeSurfer.
• Power analysis indicated that the improved segmentation accuracy significantly reduces the required sample size for detecting a 5% change in brain region volume.

Conclusions:

• The proposed MAP-QMRF classifier offers enhanced accuracy in human brain MR image segmentation.
• This advancement has implications for more efficient and statistically powerful neuroimaging studies.
• The QMRF model provides a robust framework for probabilistic modeling in medical image analysis.