Homeomorphic brain image segmentation with topological and statistical atlases.

Area of Science:

• Neuroimaging
• Medical Image Analysis
• Computational Anatomy

Background:

• Atlas-based segmentation is crucial for brain image analysis but faces challenges in modeling anatomical knowledge without bias.
• Existing methods struggle to balance anatomical accuracy with flexibility for diverse brain anatomies and image qualities.

Purpose of the Study:

• To develop a robust brain image segmentation framework incorporating topological information as a prior.
• To ensure strict topological equivalence between segmented images and atlases while minimizing reliance on statistical shape priors.

Main Methods:

• A novel segmentation framework combining topological and statistical atlases of brain anatomy.
• Utilizes tissue classification and fast marching methods to handle image variations like noise and inhomogeneities.
• Guarantees topological equivalence between the segmented image and the reference atlas.

Main Results:

• Demonstrated accuracy and robustness in segmenting simulated and real brain image data.
• The method effectively handles multiple image contrasts, noise, and anatomical variations.
• Showcased limited influence of the statistical atlas, highlighting the strength of the topological prior.

Conclusions:

• The proposed framework offers a powerful and flexible approach for brain image segmentation.
• Integrating topological priors enhances segmentation accuracy and reduces bias compared to purely statistical methods.
• The freely available algorithm facilitates further research in neuroimaging analysis.