Nonrigid registration of images with different topologies using embedded maps

Area of Science:

• Medical image analysis
• Computational anatomy
• Differential geometry

Background:

• Medical images exhibit topological changes from anatomical variation, artifacts, and pathology.
• Non-rigid registration methods struggle with topological changes due to imposed smoothness constraints.
• Existing methods often fail to accurately model or segment images with complex topological alterations.

Purpose of the Study:

• To develop a robust non-rigid registration method for medical images with topological changes.
• To enable accurate atlas-based segmentation and deformation analysis in the presence of topological variations.
• To address the limitations of current registration techniques that enforce transformation smoothness.

Main Methods:

• Treating medical images as embedded maps deforming within a Riemannian space.
• Developing a novel approach to model smooth transformations that represent topological changes.
• Utilizing a partial differential equation to describe the evolution of these transformations.

Main Results:

• Demonstrated the ability to obtain smooth transformations for images with topological changes.
• Successfully modeled complex geometric transformations arising from topological alterations.
• Validated the method using 2D brain morphometry examples.

Conclusions:

• The proposed Riemannian space approach effectively handles topological changes in medical image registration.
• This method improves accuracy in atlas-based segmentation and deformation analysis for complex anatomical variations.
• The technique offers a promising solution for analyzing medical images with non-smooth geometric transformations.