MENU

Click here to

×

Are you sure ?

Yes, do it No, cancel

Improved PET Quantification by Adaptive Denoising and Regularized Image Reconstruction: A Feasibility Study

M Namias1*, R Jeraj2 , (1) Fundacion Centro Diagnostico Nuclear, Buenos Aires, Argentina, (2) University of Wisconsin, Madison, WI

Presentations

(Monday, 7/30/2018) 4:30 PM - 6:00 PM

Room: Davidson Ballroom B

Purpose: EANM guidelines recommend harmonizing PET reconstructions using smoothing filters to bring contrast recovery coefficients (CRCmax) values within specified tolerances. However, these tolerances cause underestimation of small structure quantification. Our goal was to study the feasibility of improving harmonization targets by using adaptive filtering and/or regularized reconstructions to achieve CRCmax values closer to unity.

Methods: A novel two-stage adaptive filter (AONLM-W) was designed by replacing the hard-thresholding stage of the BM4D algorithm with an adaptive and optimized non-local means filter (AONLM). NEMA PET phantoms with a 10:1 sphere to background ratio were scanned multiple times on GE Discovery 710 (GE-D710), GE Optima 560 (GE-O560) and Siemens Biograph Truepoint TrueV (S-Bio) PET/CT scanners, using clinical acquisition times. Images were reconstructed with the VuePoint-FX (GE-D710), VuePoint-HD (GE-O560) and AW-OSEM (S-Bio) algorithms without post-filters and the regularized Q.Clear algorithm (GE-D710) with a high noise suppression factor (β=800). The VuePoint and AW-OSEM images were denoised with the BM4D and AONLM-W filters. CRCmax values were calculated as maximum sphere uptake normalized by the real activity concentration. New harmonization targets were defined using the average CRCmax and ±1.96 standard deviations across reconstructions for every sphere.

Results: Q.Clear and AONLM-W achieved similar quantitative performance in terms of average CRCmax values, with Q.Clear having higher inter-scan reproducibility (average coefficient of variation for all spheres CV(CRCmax) = 4.2% vs. 7.39%). BM4D images had higher variance (CV(CRCmax) of the smallest sphere = 20%) plus visible line and block artifacts and were discarded. New CRCmax harmonization targets were [0.69±0.15 1.05±0.12 1.10±0.04 1.11±0.06 1.07±0.06 1.06±0.06] which are closer to unity for the smaller spheres than EANM targets [0.46±0.23 0.72±0.26 0.87±0.28 0.96±0.26 1.02±0.22 1.06±0.21] for the 10, 13, 17, 22, 28 and 37 mm diameter spheres respectively.

Conclusion: EANM harmonization targets for CRCmax were improved by using regularized reconstructions or adaptive filters.

Keywords

Noise Reduction, PET, Quantitative Imaging

Taxonomy

IM- PET : Quantitative imaging/analysis

Contact Email