[Paper Review] A Nonlinear Variational Approach to Motion-Corrected Reconstruction of Density Images

Martin Burger, Jan Modersitzki+1 more|arXiv (Cornell University)|Nov 29, 2015

Medical Imaging Techniques and Applications40 references3 citations

TL;DR

This paper proposes a nonlinear variational framework for motion-corrected reconstruction of density images from dynamic, indirect measurements—particularly in dynamic PET—by modeling motion as a hyperelastic, mass-preserving deformation of an initial density. The method combines data fidelity, total variation regularization, and hyperelastic regularization with a weak diffeomorphism-based motion model, proving existence of minimizers and demonstrating improved image quality and noise suppression over conventional techniques.

ABSTRACT

The aim of this paper is to establish a nonlinear variational approach to the reconstruction of moving density images from indirect dynamic measurements. Our approach is to model the dynamics as a hyperelastic deformation of an initial density including preservation of mass. Consequently we derive a variational regularization model for the reconstruction, which - besides the usual data fidelity and total variation regularization of the images - also includes a motion constraint and a hyperelastic regularization energy. Under suitable assumptions we prove the existence of a minimizer, which relies on the concept of weak diffeomorphisms for the motion. Moreover, we study natural parameter asymptotics and regularizing properties of the variational model. Finally, we develop a computational solution method based on alternating minimization and splitting techniques, with a particular focus on dynamic PET. The potential improvements of our approach compared to conventional reconstruction techniques are investigated in appropriately designed examples.

Motivation & Objective

• To address the challenge of reconstructing moving density images from undersampled or noisy dynamic measurements, particularly in cardiac PET.
• To overcome limitations of separate time-step reconstructions, which suffer from poor signal-to-noise ratio due to low counts or undersampling.
• To develop a unified variational model that simultaneously reconstructs images and estimates motion with physical consistency.
• To ensure mass conservation and smooth, invertible deformations via a hyperelastic regularization energy and weak diffeomorphism framework.
• To establish theoretical convergence and regularizing properties of the proposed model through existence and asymptotic analysis.

Proposed method

• The method models the time-evolving density as a hyperelastic deformation of an initial density, enforcing mass conservation via the Jacobian determinant in the transformation equation $\rho^i(x) = \rho^0(y^i(x)) \cdot \det(\nabla y^i(x))$.
• A variational functional is constructed combining data fidelity to measured projections, total variation regularization for image sparsity, and hyperelastic regularization for motion smoothness.
• Motion is constrained using a weak diffeomorphism framework, ensuring a.e. approximate differentiability, positive Jacobian determinant, and global invertibility via the area formula and Lusin’s condition.
• The model incorporates a norm $\|y\|_{\operatorname{dif}^{p,q}}$ based on the deformation gradient, cofactor, and determinant to enforce physical plausibility and regularity.
• An alternating minimization algorithm is developed, splitting the functional into image and motion subproblems for efficient numerical solution.
• The approach is validated in dynamic PET with Poisson-distributed noise, using negative log-likelihood data fidelity and iterative refinement with motion correction.

Experimental results

Research questions

• RQ1 Can a nonlinear variational model that couples image reconstruction and motion estimation yield higher-quality density images than conventional separate reconstructions in dynamic PET?
• RQ2 How can mass conservation and physically plausible deformations be mathematically enforced in a variational framework for moving density images?
• RQ3 What are the theoretical properties—such as existence of minimizers and regularizing behavior—of a variational model combining data fidelity, image regularization, and hyperelastic motion energy?
• RQ4 How does the use of weak diffeomorphisms ensure invertibility and smoothness of the deformation field while allowing for large deformations?
• RQ5 What are the convergence and stability properties of the alternating minimization scheme used to solve the joint reconstruction and motion estimation problem?

Key findings

• The proposed variational model admits a minimizer under suitable assumptions, proven via the theory of weak diffeomorphisms and global invertibility conditions.
• The existence of a minimizer relies on the use of weak diffeomorphisms, which ensure a.e. approximate differentiability, positive Jacobian determinant, and global invertibility.
• The model exhibits regularizing properties, with asymptotic behavior showing convergence to a physically plausible solution as regularization parameters are adjusted.
• The computational method based on alternating minimization and splitting techniques converges consistently, avoiding artifacts common in iterative schemes that alternate reconstruction and motion estimation.
• Numerical examples demonstrate significant improvements in signal-to-noise ratio and image quality compared to conventional reconstruction, particularly in low-count dynamic PET scenarios.
• The framework is extendable to other noise models beyond Poisson, such as additive or background noise, by replacing the data fidelity term with the corresponding negative log-likelihood.