[A multicenter evaluation of seven commercial ML-EM algorithms for SPECT image reconstruction using simulation data]

Area of Science:

• Medical Imaging
• Nuclear Medicine
• Image Reconstruction

Background:

• Maximum Likelihood Expectation Maximization (ML-EM) is an alternative to filtered back projection for Single Photon Emission Computed Tomography (SPECT).
• Computational details vary between SPECT manufacturers and models, potentially affecting ML-EM algorithm performance.

Purpose of the Study:

• To investigate the characteristics of seven different ML-EM algorithms.
• To compare image quality metrics across various ML-EM implementations using simulated SPECT data.

Main Methods:

• Seven ML-EM algorithm programs (Genie, esoft, HARP-III, GMS-5500UI, Pegasys, ODYSSEY-FX, Windows-PC) were evaluated.
• Simple simulation data of a line source were reconstructed with ML-EM, varying iterations from 1 to 45.
• Image quality assessed using Full Width at Half Maximum (FWHM), Full Width at Tenth Maximum (FWTM), and total counts.

Main Results:

• Significant differences in FWHM (up to 1.5 pixels) and FWTM (no less than 2.0 pixels) were observed among algorithms at maximum iterations.
• Total counts varied in early iterations, differing from converged values based on initial parameters.
• Each ML-EM algorithm demonstrated unique reconstruction behaviors, akin to generating its own simulation image.

Conclusions:

• ML-EM SPECT algorithms exhibit distinct characteristics influenced by their computational underpinnings.
• Awareness of the specific ML-EM algorithm and its computational details is essential for comparing physical and clinical usefulness.
• Further research into algorithm-specific performance is needed for reliable SPECT image interpretation.