Acceleration of image-based resolution modelling reconstruction using an expectation maximization nested algorithm

Area of Science:

• Medical Imaging
• Computational Science
• Image Reconstruction

Background:

• Iterative reconstruction algorithms benefit from resolution models to counteract image degradation.
• Existing models face slower convergence due to integrated deconvolution processes.

Purpose of the Study:

• To evaluate a recently proposed algorithm that separates tomographic and deconvolution problems within an image-based expectation maximization (EM) framework.
• To assess the algorithm's ability to accelerate convergence and improve image resolution.

Main Methods:

• Implementation of a novel algorithm decoupling tomographic and image deconvolution problems.
• Evaluation using 2D simulations and 3D data from a high-resolution research tomograph.
• Interleaving multiple image-based EM iterations for resolution modeling with single EM iterations for tomography.

Main Results:

• Accelerated bias reduction and improved spatial resolution, particularly at high-contrast boundaries and in small regions of interest.
• Significant acceleration of convergence, allowing for up to 70% fewer tomographic iterations for small regions.
• Demonstrated effectiveness in both simulated and real-world 3D imaging data.

Conclusions:

• The proposed nested algorithm effectively accelerates convergence in iterative image reconstruction.
• Decoupling tomographic and deconvolution tasks enhances resolution recovery efficiency.
• Algorithm performance is application-dependent, requiring careful selection of nested EM iterations.