Similarity-regulation of OS-EM for accelerated SPECT reconstruction

Area of Science:

• Medical Imaging
• Nuclear Medicine
• Computational Science

Background:

• Ordered subsets expectation maximization (OS-EM) accelerates image reconstruction in single photon emission computed tomography (SPECT).
• High subset numbers improve speed but can cause artifacts in low-count projections.
• Count-regulated OS-EM (CR-OS-EM) prevents artifacts but remains slow in low-activity regions.

Purpose of the Study:

• To develop a faster image reconstruction algorithm for SPECT that overcomes the speed limitations of CR-OS-EM.
• To introduce Similarity-regulated OS-EM (SR-OS-EM) that locally adapts subset numbers based on voxel update similarity.

Main Methods:

• Proposed SR-OS-EM algorithm that dynamically adjusts the number of subsets per voxel.
• Evaluated SR-OS-EM using an image quality phantom and in vivo scans.
• Compared SR-OS-EM performance against ML-EM and CR-OS-EM.

Main Results:

• SR-OS-EM retains favorable properties of CR-OS-EM while achieving significantly higher speeds in low-activity regions.
• Reconstruction speed in low-activity regions was up to an order of magnitude faster than CR-OS-EM.
• SR-OS-EM demonstrated consistent performance across various count levels, simplifying parameter selection.

Conclusions:

• SR-OS-EM offers a substantial speed improvement over CR-OS-EM for SPECT image reconstruction.
• The algorithm's ability to adapt subset numbers locally enhances efficiency without compromising image quality.
• SR-OS-EM provides a more robust and user-friendly solution for SPECT imaging, particularly in challenging low-count scenarios.