An overview of fast convergent ordered-subsets reconstruction methods for emission tomography based on the...

Area of Science:

• Medical Imaging
• Computational Science
• Statistical Modeling

Background:

• Statistical reconstruction is crucial for emission computed tomography (ECT) but often exhibits slow convergence.
• Existing methods like OSEM are fast but non-convergent, while convergent algorithms (RAMLA, BSREM) require user-defined relaxation schedules.

Purpose of the Study:

• To develop a fast and convergent statistical reconstruction algorithm for ECT.
• To address the limitations of existing algorithms by eliminating the need for a relaxation schedule.

Main Methods:

• Proposed Complete-data OSEM (COSEM) algorithms for both ML and MAP cases.
• COSEM is derived as a form of incremental EM algorithm.
• Demonstrated COSEM performance using simulations.

Main Results:

• COSEM achieves fast convergence for both ML and MAP solutions in ECT.
• COSEM avoids the use of a relaxation schedule while maintaining convergence.
• COSEM-ML and COSEM-MAP show competitive performance compared to existing methods, with potential for further optimization.

Conclusions:

• COSEM provides an effective solution for accelerating statistical reconstruction in ECT.
• The algorithm's ability to maintain convergence without a relaxation schedule simplifies its application.
• Further modifications to COSEM can potentially overcome early-iteration speed limitations.