Multi-atlas segmentation with particle-based group-wise image registration.

Area of Science:

• Neuroimaging
• Medical Image Analysis
• Computational Biology

Background:

• Accurate brain segmentation is crucial for analyzing rodent magnetic resonance (MR) images in neuroscience research.
• Existing segmentation methods can be sensitive to image quality, variations in brain size/shape, and template bias.

Purpose of the Study:

• To develop a novel multi-atlas segmentation method for rodent brain MR images.
• To improve the robustness and accuracy of brain segmentation using a particle-guided image registration technique.

Main Methods:

• A group-wise image registration method based on particle correspondence in the volumetric domain was developed.
• The method simultaneously registers multiple images to an average particle space, creating an implicit common reference frame.
• The particle-guided registration was extended to a multi-atlas segmentation approach, incorporating template labels as constraints.

Main Results:

• The particle-guided registration method demonstrated robustness with low signal-to-noise ratio images and variations in rodent brain development.
• The novel multi-atlas segmentation method achieved higher accuracy and stability compared to traditional pair-wise registration.
• The approach mitigated potential bias associated with single-template segmentation.

Conclusions:

• Particle-guided image registration offers a robust solution for segmenting rodent brain MR images, particularly in challenging scenarios.
• The developed multi-atlas segmentation method significantly improves segmentation accuracy and reliability.
• This technique provides a valuable tool for quantitative analysis in preclinical neuroimaging studies.