Brain tissue segmentation using improved kernelized rough-fuzzy C-means with spatio-contextual information from MRI.

Area of Science:

• Medical Imaging and Image Analysis
• Computational Intelligence
• Biomedical Engineering

Background:

• Accurate brain tissue segmentation from MRI is critical for diagnosing neurological abnormalities.
• Conventional segmentation methods struggle with imprecise, ambiguous, overlapping, and non-linearly separable brain tissues, as well as noise and artifacts.

Purpose of the Study:

• To propose a robust kernelized rough fuzzy C-means clustering with spatial constraints (KRFCMSC) method for improved brain tissue segmentation from MRI.
• To address the challenges of ambiguity, indiscernibility, vagueness, overlapping regions, and noise in brain MRI segmentation.

Main Methods:

• Developed a novel KRFCMSC algorithm integrating fuzzy set theory, rough set theory, kernel trick, and spatial constraints for pixel clustering.
• Incorporated spatio-contextual information from neighboring pixels to mitigate the impact of noisy pixels.
• Validated the method on real and synthetic brain MRI datasets (Brainweb, IBSR) with and without added noise.

Main Results:

• The proposed KRFCMSC method demonstrated superior and robust performance compared to five other clustering-based segmentation techniques.
• Evaluated using multiple supervised and unsupervised validity indices (accuracy, precision, recall, kappa, Jaccard, dice, kernelized Xie-Beni index).
• Paired t-test results confirmed statistically significant improvements in segmentation accuracy.

Conclusions:

• The KRFCMSC method effectively handles complex challenges in brain MRI segmentation, outperforming existing state-of-the-art approaches.
• The integration of rough and fuzzy sets, kernel trick, and spatial constraints provides a robust solution for accurate brain tissue segmentation, even in the presence of noise.