Dynamic FDG PET imaging using direct 4D parametric reconstruction in cancer patients

Objectives To investigate applicability of direct reconstruction of parametric images from dynamic oncologic FDG PET data for the purpose of improved tumor detection.

Methods A direct 4D iterative expectation-maximization (EM) image reconstruction algorithm, involving direct estimation of Patlak slope/intercept parameters from the sinogram data, was applied to dynamic oncologic FDG studies. Four 60min FDG studies (6x10s,3x20s,2x90s,2x150s,2x300s,4x600s), performed on patients with metastatic renal cell carcinoma (total of 7 tumors), were considered. Three types of images were obtained: (i) SUV images based on the last 10min frame, (ii) standard Patlak parametric images as obtained from independently reconstructed frames, (iii) directly reconstructed Patlak parametric images. To estimate the blood input function, the left ventricle or the descending aorta were used (depending on whether the heart was in the field of view). Tumor-to-background contrast-to-noise ratios (T/B CNR) were optimized across varying iterations for each reconstruction method, and compared using paired t-test.

Results Compared to SUV images, standard Patlak-slope parametric imaging improved T/B CNR (p<0.05) by an average of 20%, as it reduced blood and free-FDG background contributions; however it generated substantial increases in noise in the images. By contrast, directly 4D reconstructed parametric images produced significantly lowered noise levels compared to standard parametric images. As a result, 4D parametric imaging resulted in significant improvements (p<0.005) in T/B CNR when compared to both (i) SUV and (ii) standard parametric images, by averages of 67% and 39%, respectively.

Conclusions Direct 4D parametric imaging as applied to dynamic oncologic FDG imaging is a very promising method of enhancing tumor contrast while controlling image noise as one switches from the SUV FDG method to quantitative parametric imaging