Multi-Object Model-based Multi-Atlas Segmentation for Rodent Brains using Dense Discrete Correspondences.

Area of Science:

• Neuroimaging
• Computational Anatomy
• Medical Image Analysis

Background:

• Accurate delineation of rodent brain structures is difficult due to low contrast and closely interfacing organs.
• Atlas-based segmentation is common but can be limited by registration errors, especially for in vivo MRI.
• Existing multi-atlas methods improve robustness but still face challenges with accuracy and artifacts.

Purpose of the Study:

• To develop a more accurate and robust atlas-based segmentation method for rodent brain structures.
• To overcome limitations of current segmentation techniques, particularly for in vivo MRI data.
• To improve the delineation of multiple cortical and subcortical brain organs.

Main Methods:

• Proposed a multi-object, model-based, multi-atlas segmentation approach.
• Established spatial correspondences using dense pseudo-landmark particles across atlases.
• Built a multi-object point distribution model to capture inter- and intra-subject brain variation.
• Obtained segmentation by fitting the model to subject images followed by label fusion.

Main Results:

• The proposed method demonstrated superior accuracy compared to existing segmentation techniques.
• Achieved greater accuracy than the widely used ANTs registration tool.
• Showcased improved robustness in segmenting rodent brain structures, including challenging low-contrast areas.

Conclusions:

• The multi-object, model-based, multi-atlas segmentation method significantly enhances rodent brain structure delineation accuracy.
• This novel approach offers improved robustness against image artifacts and registration errors.
• The method provides a valuable tool for neuroimaging research requiring precise brain segmentation.