PT - JOURNAL ARTICLE
AU - Moody, Jonathan
AU - Lee, Benjamin
AU - Ficaro, Edward
TI - Comparison of ROI quantification for three PET reconstruction methods
DP - 2009 May 01
TA - Journal of Nuclear Medicine
PG - 1509--1509
VI - 50
IP - supplement 2
4099 - http://jnm.snmjournals.org/content/50/supplement_2/1509.short
4100 - http://jnm.snmjournals.org/content/50/supplement_2/1509.full
SO - J Nucl Med2009 May 01; 50
AB - 1509 Objectives We evaluated PET region-of-interest (ROI) quantificationby comparing the bias variance tradeoff for three PET reconstructionmethods: standard OSEM, quadratically penalized weighted least squares(PWLS), and a wavelet-based iterative shrinkage-thresholding (IST)method. Methods The IST algorithm was implemented for emission reconstructionusing the dual-tree complex wavelet transform. PET simulations were performed on a digital NEMA NU2-2001 phantom with six spheres of diameter (10,13,17,22,28,37)mm and 4:1 sphere:background activity concentration ratio. Sinograms were simulated with attenuation, nonuniform detector efficiency, randomcoincidences (10%) and Poisson noise with approximately 2e6 total counts. The noisy sinograms were reconstructed by each method. OSEM reconstructions were post-smoothed with a 6mm Gaussian kernel. Sphere activity was quantified using the known sphere boundaries. For each sphere the mean-squared error (MSE) relative to the true sphere activity was averaged over 200 noise realizations, and bias variancetradeoff curves were calculated per sphere. Results Global MSE of IST images was lower than PWLS by 21% and OSEMby 8%. The amount of regularization that minimized the MSE per spherevaried by sphere diameter. Post-smoothed OSEM produced the lowestvariance for 4 of 6 spheres, but significantly higher bias for allspheres (40-50%). The IST method produced the lowest bias for(13,17,28,37)mm spheres by 2-13%, the lowest variance for (22,37)mmspheres by 14-24%, and variance within 6% of lowest for other spheresgreater than 10mm. Conclusions The wavelet-based IST PET reconstruction method mayprovide reduced bias and improved noise properties compared toconventional methods, leading to improved ROI quantification.