GUBS: Graph-Based Unsupervised Brain Segmentation in MRI Images.

Area of Science:

• Medical Imaging
• Computational Neuroscience
• Image Analysis

Background:

• Brain segmentation in MRI is crucial for analyzing pathologies and volumes.
• Existing methods often require time-intensive and costly manual labeling.
• There is a need for automated, unsupervised segmentation techniques.

Purpose of the Study:

• To introduce Graph-based Unsupervised Brain Segmentation (GUBS), a novel algorithm for segmenting 3D MRI images.
• To segment images into distinct regions: brain, non-brain tissues, and background.
• To provide an unsupervised alternative to labor-intensive, supervised segmentation methods.

Main Methods:

• GUBS constructs an adjacency graph from preprocessed 3D MRI data.
• It computes the Minimum Spanning Tree (MST) and uses domain knowledge to identify representative points.
• Subgraph computation on the MST isolates brain, non-brain, and background components.

Main Results:

• GUBS successfully segmented 3D T1-weighted MRI images from multiple datasets.
• The method achieved performance comparable to state-of-the-art supervised techniques.
• Key advantage: GUBS does not require labeled training data.

Conclusions:

• GUBS offers an effective unsupervised approach for brain MRI segmentation.
• Its performance is on par with supervised methods, highlighting its potential.
• The elimination of manual labeling significantly reduces time and cost in brain image analysis.