Computional cost of nonrigid registration algorithms based on fluid dynamics

Area of Science:

• Medical image analysis
• Computational fluid dynamics
• Image registration

Background:

• Nonrigid registration is crucial for medical image analysis, enabling accurate comparisons of anatomical structures.
• Fluid dynamics models offer robust solutions for nonrigid registration, even with significant tissue deformation.
• Current fluid dynamics registration methods are computationally intensive, limiting their clinical applicability.

Purpose of the Study:

• To investigate and compare different computational approaches for solving the fluid dynamics partial differential equations used in nonrigid registration.
• To evaluate the trade-offs between computational cost and registration accuracy for various solvers.
• To identify the most efficient and accurate solver for fluid dynamics-based nonrigid registration.

Main Methods:

• Exploration of various numerical solvers for the partial differential equations governing fluid dynamics in image registration.
• Comparative analysis of computational costs (time and resources) associated with different solvers.
• Assessment of registration accuracy using quantitative metrics on datasets with varying degrees of deformation.

Main Results:

• Identified relaxation methods as the most promising approach for fluid dynamics-based nonrigid registration.
• Demonstrated that focusing updates on deformation hotspots significantly reduces computational cost per iteration.
• Achieved a favorable balance between computational efficiency and high registration accuracy with optimized relaxation solvers.

Conclusions:

• Relaxation-based solvers represent the most effective strategy for accelerating fluid dynamics nonrigid registration.
• Optimizing solver updates can substantially decrease the time required for accurate image registration.
• This work provides a pathway towards more practical and efficient application of fluid dynamics in medical image analysis.