Investigation of the preconditioner-parameter in the preconditioned Chambolle-Pock algorithm applied to...

Area of Science:

• Medical Imaging
• Computational Imaging
• Optimization Algorithms

Background:

• Optimization-based image reconstruction is vital in medical imaging.
• The Chambolle-Pock (CP) algorithm solves convex optimization problems in image reconstruction.
• The preconditioned CP (PCP) algorithm offers improved convergence over the ordinary CP (OCP) algorithm.

Purpose of the Study:

• Investigate the impact of the preconditioner-parameter on PCP algorithm convergence for TVDM image reconstruction.
• Determine if a fixed preconditioner-parameter (e.g., 1) is optimal across different imaging modalities and conditions.
• Provide guidance on optimizing PCP algorithm performance through adaptive parameter tuning.

Main Methods:

• Applied the preconditioned Chambolle-Pock (PCP) algorithm to TV constrained, data-divergence minimization (TVDM) image reconstruction.
• Evaluated the algorithm's convergence rate across various preconditioner-parameter values (0-2).
• Conducted studies using 2D computed tomography (CT) with Shepp-Logan and FORBILD phantoms, and 3D electron paramagnetic resonance imaging (EPRI) with simulated and physical phantoms.

Main Results:

• The optimal preconditioner-parameter for the PCP algorithm is dependent on specific imaging conditions.
• A fixed preconditioner-parameter value of 1 does not consistently yield the fastest convergence rate.
• Varying the preconditioner-parameter significantly impacts the convergence speed of the PCP algorithm in both CT and EPRI.

Conclusions:

• Adaptive tuning of the preconditioner-parameter is essential for maximizing the convergence rate of the PCP algorithm.
• The optimal preconditioner-parameter must be determined empirically for each specific imaging application.
• This adaptive approach enhances the efficiency of optimization-based image reconstruction methods.