[Current state and prospect of image reconstruction methods for PET]

[Comparison of OS-EM reconstruction algorithms among different processors using a digital phantom dedicated for SPECT data evaluation]

OBJECTIVE

In the OS-EM method, reconfigured images may be different among reconstruction image processors because the programs are highly arbitrary. The purpose of this study was to evaluate the difference in OS-EM algorithms among four different image processors using a digital phantom dedicated for SPECT data evaluation. The image processors used were GMS-5500A/PI (Toshiba), GENIE Xeleris (GE), e.soft (Siemens), and Odyssey FX (Shimadzu).

METHODS

Multiple images were reconstructed with OS-EM fluctuating the number of subsets and iterations. A region of interest (ROI) was placed on each image. The average counts, contrast, root mean square uncertainty (%RMSU), and normalized mean squared error (NMSE) of each ROI were calculated and compared with those of different image processors.

RESULTS

There was no significant difference in contrast among the algorithms. However, the average counts and (%RMSU were significantly different among algorithms as the number of updates increased. In addition, the minimums of NMSE were also different among algorithms.

CONCLUSION

In the OS-EM method, careful evaluation is necessary when using multiple image processors in research studies on the standardization of nuclear medicine imaging or in clinical applications.

[Quantitative evaluation of block-iterative reconstruction image in PET: comparison of the dynamic RAMLA algorithm and OSEM algorithm]

PURPOSE

Iterative reconstruction has been successfully used in whole-body PET imaging because of reductions in noise and scanning time. However, there are plural algorithms for image reconstruction such as OSEM, RAMLA and Dynamic RAMLA. Dynamic RAMLA (DRAMA) is an iterative algorithm similar to RAMLA, but the relaxation parameter is controlled in such a way that the propagation of noise from projection data to the reconstructed image. The purpose of this study was to investigate differences in the DRAMA and OSEM algorithms, in terms of real space and frequency space.

METHOD

A whole-body torso phantom (NEMA image-quality phantom) filled with F-18 was scanned with a dedicated PET scanner. The smallest four spheres, with internal diameters of 10, 13, 17, and 22 mm, were filled with F-18 at an 8:1 concentration with respect to the background. The two largest spheres, with diameters of 28 and 37 mm, were not filled with water (empty). The emission scan for a 50 cm axial range varied from 5-20 minutes, to examine the effect of the count statistics on the quality of the reconstructed images. The images were reconstructed with OSEM by changing the number of subsets from 1 to 128 in each study. As to the number of iterations, both iterative reconstruction algorithms were one. As a reference standard, images with maximum counts (60 minutes) were reconstructed with a filtered-back projection method. The quality of the reconstructed images was evaluated in terms of contrast, background variability, and two-dimensional power spectrum analysis.

RESULTS

As for the real space results, OSEM with more than 32 subsets decreased contrast because of images with checkerboard noise. The DRAMA algorithm provided stable contrast even when count statistics were poor. A two-dimensional power spectrum analysis also revealed that the OSEM algorithm enhanced noise components in reconstruction with more than 32 subsets.

CONCLUSION

Our preliminary data suggested that the OSEM algorithm requires few iterations with a small number of subsets for whole-body imaging. However, the DRAMA algorithm provides a reasonable signal-to-noise ratio with satisfactory spatial resolution even with one iteration. Especially in clinical whole-body PET, DRAMA was most useful because of its fast convergence and small computer burden.