Robust kernelized local information fuzzy C-means clustering for brain magnetic resonance image segmentation

Area of Science:

• Medical Imaging
• Computer Vision
• Biomedical Engineering

Background:

• Brain tissue segmentation from magnetic resonance (MR) images is crucial for clinical applications.
• Segmentation is complicated by noise, grayscale inhomogeneity, and image artifacts.
• Existing methods may struggle with accuracy and robustness in challenging conditions.

Purpose of the Study:

• To propose a novel algorithm, the robust kernelized local information fuzzy C-means clustering (RKLIFCM), for improved brain MR image segmentation.
• To enhance segmentation homogeneity by incorporating both grayscale and spatial local information.
• To improve robustness against noise and image artifacts while preserving image details.

Main Methods:

• Developed the RKLIFCM algorithm, integrating local grayscale and spatial information.
• Utilized a Gaussian radial basis kernel function as a distance metric, replacing the standard Euclidean distance.
• Implemented the algorithm to run on image histograms for high-speed processing.

Main Results:

• The RKLIFCM algorithm demonstrated efficient utilization of local information and robustness to noise.
• Achieved higher segmentation accuracy compared to 7 other soft clustering algorithms.
• Showcased preservation of image details and high-speed computation due to histogram-based processing.

Conclusions:

• The proposed RKLIFCM algorithm effectively overcomes noise influence in brain MR image segmentation.
• Achieves superior segmentation accuracy with low computational complexity.
• Presents a robust and efficient solution for clinical brain MR image analysis.