A variational formulation for discrete registration.

Area of Science:

• Medical image analysis
• Computational anatomy
• Computer vision

Background:

• Deformable registration is crucial for aligning medical images.
• Existing methods face challenges in computational efficiency and accuracy.
• Variational formulations offer a promising framework for registration.

Purpose of the Study:

• To present a novel variational formulation for discrete deformable registration.
• To demonstrate the equivalence between finite difference and MRF formulations.
• To develop an efficient computational solution using the finite element method.

Main Methods:

• Developed a convex energy functional with diffusion regularization for deformable registration.
• Showed equivalence between finite difference (FD) and Gaussian Markov random field (MRF) formulations.
• Implemented a finite element method (FEM) for efficient minimization of the energy functional.

Main Results:

• The proposed variational formulation is equivalent to the random walker method.
• FEM provides a computationally efficient solution for the variational problem.
• The method achieved competitive performance against 14 other deformable registration techniques on the CUMC12 MRI dataset.

Conclusions:

• The novel variational formulation provides an effective and efficient approach to deformable registration.
• The FEM-based solution enhances computational feasibility for complex registration tasks.
• This method shows strong potential for applications in medical image analysis and comparison.