Effect of Different Reconstruction Methods on Quantitative Analysis of Cardiac Function Using FDG Cardiac Gated PET/MR

Purpose: To evaluate the effect of different methods of reconstructions and number of iterations on the quantitative analysis of cardiac function using cardiac gated PET. Materials and Methods: 15 patients underwent 18F-FDG cardiac PET/MR examination, PET and MRI data were collected simultaneously. The filtered back projection, the ordered subsets expectation maximization (OSEM) and the OSEM based on the point spread function (PSF-OSEM) were combined with ECG-gating to reconstruct the images. The number of iterations was defined as 1 to 8 times. PET/MR cardiac software of SIEMENS Singo via workstation were used for PET and cine image processing to obtain the left ventricular ejection fraction (LVEF) and the percentage of left ventricular wall thickness of AHA 1~16 segment from end-diastolic to end-systolic stage(Thickening). ANOVA was used to compare the difference among PET images obtained by different reconstruction methods. Pearson correlation analysis was used for the analysis of the correlation between PET and cine images.

Results: There was no significant difference in the LVEF_PET obtained by different reconstruction methods (P>0.5), which were in good correlation (r=0.788~0.973, P<0.05). Two iterations OSEM method achieved the highest correlation (r=0.973), which was overestimated by (0.08±17.07)% compared to LVEF_MR. Only between AHA 2 and 3 segment thickening_PET obtained by different reconstruction methods was found significant difference(F=3.69,2.34,P<0.05) There was good correlation with AHA 1 and 4~16 thickening_MR(r=0.45~0.791, P<0.05). Also 2 iterations OSEM method had the highest correlation (r=0.791), overestimated by(9.08±22.69)% compared with thickening_MR.

Conclusions: There was no significant difference between PET images of the LVEF_PET, AHA 1 and 4~16 segments using different reconstruction methods. Cardiac PET images of LVEF_PET, AHA 1 and 4~16 segments are more accurate by using 2 iteration OSEM.