Robust generative asymmetric GMM for brain MR image segmentation.

Area of Science:

• Medical Imaging
• Computational Neuroscience
• Image Processing

Background:

• Accurate brain tissue segmentation from MR images is crucial for neurological studies.
• Existing Gaussian Mixture Model (GMM) methods struggle with noise and intensity inhomogeneity.
• Improved segmentation accuracy is needed for reliable analysis of brain MR images.

Purpose of the Study:

• To present a Robust Generative Asymmetric GMM (RGAGMM) for enhanced brain MR image segmentation.
• To simultaneously address intensity inhomogeneity and noise in MR images.
• To improve the accuracy of unsupervised brain MR image segmentation.

Main Methods:

• Developed a spatial constrained asymmetric model using an asymmetric distribution.
• Incorporated pseudo-likelihoods and bias field estimation into the log-likelihood.
• Utilized an expectation maximization (EM) algorithm for simultaneous segmentation and inhomogeneity correction.

Main Results:

• Demonstrated algorithm performance on synthetic and clinical 3T brain MR images.
• Showcased superior handling of noise and intensity inhomogeneity compared to state-of-the-art methods.
• Achieved over 5% improvement in segmentation accuracy on benchmark datasets (IBSR, BrainWeb).

Conclusions:

• The RGAGMM algorithm offers efficient simultaneous segmentation and intensity inhomogeneity correction.
• The method effectively overcomes noise and inhomogeneity influences in brain MR images.
• RGAGMM provides a flexible and accurate approach for brain MR image analysis.