Symmetric log-domain diffeomorphic Registration: a demons-based approach

Area of Science:

• Computational anatomy
• Medical image analysis
• Non-linear registration

Background:

• Modern morphometric studies rely on non-linear image registration for anatomical comparison and group analysis.
• Log-Euclidean approaches simplify statistical computations on spatial transformations, advancing computational anatomy tools.

Purpose of the Study:

• To propose a novel non-linear registration algorithm optimized for log-Euclidean statistics on diffeomorphisms.
• To ensure invertibility and direct usability of transformations for statistical analysis.

Main Methods:

• The algorithm operates entirely in the log-domain using a stationary velocity field.
• It employs an alternating optimization strategy, inspired by Thirion's demons algorithm.
• Symmetry with respect to input image order is maintained.

Main Results:

• The proposed algorithm guarantees invertible deformations and provides direct access to the true inverse transformation.
• It outperforms the demons and diffeomorphic demons algorithms in transformation accuracy.
• The method demonstrates computational efficiency comparable to existing algorithms.

Conclusions:

• The novel log-domain non-linear registration algorithm facilitates accurate and efficient computational anatomy.
• It enables direct application of log-Euclidean statistics on diffeomorphic transformations.
• This approach advances the field by improving accuracy and maintaining computational efficiency.