Subset-dependent relaxation in block-iterative algorithms for image reconstruction in emission tomography

Area of Science:

• Medical Imaging
• Computational Physics
• Image Reconstruction

Background:

• Iterative algorithms are crucial for image reconstruction in Positron Emission Tomography (PET).
• Conventional Ordered Subset Expectation Maximization (OS-EM) algorithms can suffer from noise amplification and slow convergence.
• Controlling noise propagation and ensuring convergence are key challenges in PET image reconstruction.

Purpose of the Study:

• To introduce a novel row-action maximum likelihood algorithm (RAMLA) with a dynamic relaxation parameter for improved PET image reconstruction.
• To develop a dynamic RAMLA (DRAMA) and a dynamic OS-EM (DOSEM) to mitigate noise propagation and enhance reconstruction quality.
• To evaluate the performance of DRAMA and DOSEM in terms of signal-to-noise ratio, spatial resolution, and convergence speed.

Main Methods:

• A subset-dependent relaxation parameter, lambda(k)(q), was formulated to control noise propagation independently of data access order.
• The dynamic RAMLA (DRAMA) was proposed as a special case of dynamic OS-EM (DOSEM) with Msub = M (number of angular views).
• Simulation studies were conducted to assess the performance of DRAMA and DOSEM compared to conventional methods.

Main Results:

• DRAMA demonstrated a substantial independence of noise propagation from subset access order.
• The proposed DRAMA achieved a good signal-to-noise ratio and satisfactory spatial resolution in few iterations for 2D PET reconstruction.
• DOSEM allowed for a larger number of subsets (OS level) without compromising signal-to-noise ratio, with DRAMA offering the fastest convergence and lowest computational load.

Conclusions:

• The dynamic RAMLA (DRAMA) effectively controls noise propagation in PET image reconstruction.
• DRAMA provides a superior balance of signal-to-noise ratio, spatial resolution, and computational efficiency.
• The developed algorithms (DRAMA and DOSEM) represent significant advancements for iterative image reconstruction in PET imaging.