Markov Random Field-based Fitting of a Subdivision-based Geometric Atlas

Area of Science:

• Medical image analysis
• Computational anatomy
• Neuroimaging

Background:

• Accurate labeling of complex anatomical structures is crucial for spatial analysis and data comparison across images.
• Geometric atlases, constructed from deformable meshes, aid in anatomical labeling but automated fitting remains challenging.
• Subdivision meshes offer a compact representation for multi-resolution, object-specific mesh structures.

Purpose of the Study:

• To develop a novel automated method for fitting a subdivision mesh-based geometric atlas to anatomical structures.
• To address the challenge of landmark matching concurrently with atlas fitting.
• To improve the accuracy and efficiency of anatomical segmentation in complex biological images.

Main Methods:

• Proposed a Markov Random Field (MRF)-based method for fitting a planar, multi-part subdivision mesh to anatomical data.
• Determined optimal control point locations for precise atlas fitting.
• Integrated landmark matching into atlas fitting using a single graphical model for pose-invariant constraints.

Main Results:

• Demonstrated the method's effectiveness on segmenting mouse brain regions in gene expression images.
• Achieved promising results in handling significant intensity and shape variability.
• Showcased improved performance compared to manual annotations and existing methods.

Conclusions:

• The novel MRF-based method provides an effective solution for automated anatomical atlas fitting.
• The integrated approach successfully handles landmark matching and complex deformations.
• This technique shows significant potential for advancing spatial analysis in biological imaging.