Optimization of Ordered Subset Expectation Maximization Parameters for Image Reconstruction in Tc-99m Methoxyisobutylisonitrile Myocardial Perfusion SPECT and Comparison with Corresponding Filtered Back Projection-Reconstructed Images

Abstract

Purpose of the study: To establish the most appropriate ordered subset expectation maximization (OSEM) parameters for image reconstruction in Tc-99m methoxyisobutylisonitrile (MIBI) myocardial perfusion SPECT (MPS) and comparison with corresponding filtered back projection (FBP)-reconstructed images.

Methods: A total of 99 stress-rest MPS studies (47 normal and 52 abnormal) were retrospectively analyzed using 16 different combinations of iterations and subsets. Images were reconstructed both with and without postreconstruction Butterworth filter (cutoff frequency and order for stress: 0.4 and 10 and for rest: 0.52 and 5, respectively) for each combination. A total of 3168 images were evaluated qualitatively by two nuclear medicine physicians on a scoring scale of 1-4. Best visual quality image iteration-subset combination was determined for each patient both with and without Butterworth filter and was further compared with FBP-reconstructed image. The interobserver agreement was obtained using kappa statistics.

Results: The best quality images were obtained using a combination of four iterations and six subsets for both with and without Butterworth filter. The value of kappa for interobserver agreement for OSEM images with Butterworth filter was 0.570 and for OSEM images without Butterworth filter was 0.857. On comparison, FBP images were better than OSEM-reconstructed images without Butterworth filter (P < 0.0001 calculated using Fisher's exact test) with substantial agreement (kappa = 0.628). However, OSEM-reconstructed images with Butterworth filter were better than FBP images and showed moderate agreement (kappa = 0.486).

Conclusion: The most appropriate OSEM reconstruction parameter in Tc-99m MIBI MPS is 4-iteration and 6-subset combination. FBP-reconstructed images were better than the images reconstructed with OSEM without postreconstruction Butterworth filter. However, OSEM-reconstructed image with Butterworth filter was better than FBP images.

Keywords: Filtered back projection; image reconstruction; myocardial perfusion SPECT; ordered subset expectation maximization.

Figure 1

Graph showing the frequencies of the iteration–subset combination with best image quality grading in images of 99 patients studies using ordered subset expectation maximization without postreconstruction Butterworth filter

Figure 2

Graph showing the frequencies of the iteration–subset combination with best image quality grading in images of 99 patients studies using ordered subset expectation maximization with postreconstruction Butterworth filter

Figure 3

Comparison of the iteration–subset combination with best image quality on ordered subset expectation maximization without postreconstruction filter and corresponding filtered back projection images for both the observers

Figure 4

Comparison of the iteration–subset combination with best image quality on ordered subset expectation maximization with postreconstruction filter and corresponding filtered back projection images for both the observers

Figure 5

Images of different reconstruction settings for comparison. (a) Normal myocardial perfusion SPECT study reconstructed using ordered subset expectation maximization without filter by iteration–subset combination of (i) 2-2, (ii) 4-6, and (iii) 8-16. (b) Normal myocardial perfusion SPECT study reconstructed using ordered subset expectation maximization with filter by iteration–subset combination of (i) 2-2, (ii) 4-6, and (iii) 8-16. (c) Comparison of best images obtained by ordered subset expectation maximization reconstruction without filter (i) and with filter (ii) with the corresponding filtered back projection-reconstructed image (iii) in normal myocardial perfusion SPECT study. Images obtained by ordered subset expectation maximization reconstruction with filter (ii in Figure 5c) were found to be the best in quality and most appropriate for processing myocardial perfusion SPECT studies