Automated segmentation of human brain MR images using a multi-agent approach

Area of Science:

• Computer Vision
• Artificial Intelligence
• Medical Imaging Analysis

Background:

• Image interpretation is a complex, distributed process requiring multiple information processing steps.
• Extracting symbolic labels from radiometric data under spatial constraints is challenging.
• Existing methods struggle with noise, non-uniformity, and partial volume effects in medical images.

Purpose of the Study:

• To propose a novel framework for image interpretation using situated cooperative agents.
• To dynamically adapt agent behaviors based on image position, relationships, and radiometric data.
• To improve radiometry-based tissue interpretation, particularly for magnetic resonance imaging (MRI) brain scans.

Main Methods:

• Utilizing situated cooperative agents with dynamically adapted behaviors.
• Employing a coarse-to-fine strategy for incremental interpretation refinement.
• Gathering and sharing information via qualitative maps for agent collaboration.
• Estimating tissue-intensity distribution models to handle noise and non-uniformity.

Main Results:

• Achieved segmentation accuracy of approximately 0.84% compared to the reference.
• Demonstrated the ability to segment a complete volume in under 5 minutes.
• Showcased the framework's effectiveness in handling noise, intensity non-uniformity, and partial volume effects.

Conclusions:

• The situated cooperative agent framework offers a promising approach for efficient and accurate image interpretation.
• The method shows significant potential for applications in magnetic resonance imaging (MRI) brain scan analysis.
• This approach effectively manages the complexities of radiometry-based tissue interpretation.