3D geometric split-merge segmentation of brain MRI datasets.

Area of Science:

• Medical Imaging
• Computer Vision
• Biomedical Engineering

Background:

• Magnetic Resonance Imaging (MRI) is crucial for visualizing anatomical structures.
• Accurate segmentation of MRI volumes is essential for quantitative analysis and diagnosis.
• Existing segmentation methods struggle with complex geometrical shapes and noisy datasets.

Purpose of the Study:

• To propose a novel MRI volume segmentation method, Adaptive Geometric Split Merge (AGSM).
• To segment complex geometrical shapes composed of homogeneous 3D regions in MRI data.
• To improve segmentation performance, especially in noisy brain MRI datasets.

Main Methods:

• Developed the Adaptive Geometric Split Merge (AGSM) segmentation technique.
• Implemented adaptive volume splitting with multiple strategies and a maximal homogeneity axis.
• Introduced region merging criteria to refine segmentation boundaries.
• Applied the method to brain MRI datasets for hard segmentation (voxel-to-tissue type assignment).

Main Results:

• AGSM successfully segments complex geometrical shapes in 3D MRI volumes.
• The method demonstrates improved segmentation performance on noisy brain MRI datasets.
• AGSM achieves superior results compared to state-of-the-art methods.
• The segmentation procedure does not require prior training data.

Conclusions:

• AGSM is an effective and robust method for MRI volume segmentation.
• The technique is particularly advantageous for segmenting noisy medical imaging data.
• AGSM offers a promising alternative for brain MRI analysis and other medical imaging applications.