A Modified Brain MR Image Segmentation and Bias Field Estimation Model Based on Local and Global Information

Area of Science:

• Medical Imaging
• Image Processing
• Computational Neuroscience

Background:

• Brain magnetic resonance (MR) images often suffer from noise and intensity inhomogeneity (bias) due to radio frequency coil non-uniformity and eddy currents.
• This noise and bias severely degrade the accuracy of image segmentation, hindering further analysis.
• Traditional segmentation methods struggle to overcome these challenges effectively.

Purpose of the Study:

• To propose a modified model for brain MR image segmentation and bias field estimation.
• To enhance segmentation accuracy by effectively addressing noise and intensity inhomogeneity.
• To develop a model that automatically adjusts the balance between local and global image information.

Main Methods:

• A modified segmentation and bias field estimation model integrating local and global information was developed.
• Local constraints were constructed using image neighborhood information within a Gaussian kernel mapping space.
• Nonlocal spatial information was introduced for complete regularization, with automatic weighting adjustment based on local image characteristics.
• Bias field information was coupled with the model to simultaneously reduce noise and estimate the bias field.

Main Results:

• The proposed algorithm demonstrated strong robustness against noise in brain MR images.
• The model effectively estimated and corrected the bias field, improving image uniformity.
• Experimental results showed superior segmentation accuracy compared to traditional methods.

Conclusions:

• The developed model successfully addresses the limitations of traditional methods for brain MR image segmentation.
• Integrating local and global information with coupled bias field estimation leads to more accurate and robust segmentation.
• This approach offers a significant improvement for analyzing noisy and biased brain MR images.