RT Journal Article
SR Electronic
T1 Three-fold reduction in pediatric PET dose by advanced image reconstruction without loss of diagnostic and quantitative performance
JF Journal of Nuclear Medicine
JO J Nucl Med
FD Society of Nuclear Medicine
SP 583
OP 583
VO 59
IS supplement 1
A1 Schmidtlein, Charles
A1 Pandit-Taskar, Neeta
A1 Haggstrom, Ida
A1 Osborne, Joseph
A1 Tang, Xinhuang
A1 Xu, Yuesheng
A1 Chen, Baiyu
A1 Zanzonico, Pat
A1 Dauer, Lawrence
A1 Krol, Andrzej
YR 2018
UL http://jnm.snmjournals.org/content/59/supplement_1/583.abstract
AB 583Aim: Radiation dose to pediatric patients due to diagnostic imaging studies including nuclear medicine is an increasing concern for patients, their families, referring physicians, and nuclear medicine professionals. Consequently, efforts to reduce pediatric patients’ radiation dose from nuclear medicine studies, without compromising the diagnostic content, is an important objective for nuclear medicine services. Objectives: We have developed and investigated the performance of a novel low-dose PET image reconstruction algorithm, a relaxed ordered-subset version of the proximity gradient algorithm (OS-PGA) using sparse representation regularization (SRR), for the purpose of reducing the injected activity, and thus the radiation dose to pediatric patients undergoing a PET/CT exam without compromising its diagnostic and quantitative performance compared to the standard-of-care (SOC) approach. Methods: Six pediatric [18F]-FDG PET/CT imaging studies were performed on GE D690 & D710 PET/CT (3 min/bed position over the torso) at 1 hour post injection with an injected activity based on an adult activity of 444 MBq FDG (1.7m2 BSA), adjusted using the pediatric-to-adult patients’ body surface area ratios. The data were subsequently extracted from list-mode to create 3-fold reduced-count PET projection data, equivalent to, 1 min/bed (torso). A total of 18 regions of high physiological uptake were investigated in the images reconstructed using SOC approach. The SRR methods are used to penalize the maximum likelihood estimate for Poisson noise in a PET data model, where fixed-point methods are used to solve the ensuing optimization problem (OS-SRR-PGA). The SOC approach applies the ordered-subsets expectation maximization (OSEM) algorithm. Both SOC and SRR algorithms reconstruct a 700 mm field-of-view (FOV) and use time-of-flight information (600 ps) and point-spread-function correction (using SharpIR, GE). The OS-SRR-PGA reconstruction uses 10/24 iterations/subsets with a 256x256 matrix and data-dependent penalty weights. SOC-OSEM uses 2/16 iterations/subsets with a 128x128 matrix, and axial and transaxial post-filtering with 3-point smoothing (Heavy), and Gaussian 6.4-mm FWHM filters, respectively. The image quality of the PET images obtained with our low-dose algorithm was evaluated based on two tasks: (i) lesion detection; and (ii) quantitative accuracy. Volumes-of-interest (VOIs) were estimated using a threshold of 2σ above the background. They were used to compare the tumor contrast-to-noise ratio (CNR) of both full-count (SOC) and reduced-count (OS-SRR-PGA) images. The subjective image quality was evaluated by physicians who assessed image artifacts, possible false-positive findings, the overall diagnostic image quality, and confidence. Results: The OS-SRR-PGA converged at a similar or faster rate as its unpenalized counterpart, row action maximum likelihood algorithm (i.e., RAMLA) and did not become stuck at a limit cycle. The clinically useful PET images were reconstructed in the timeframes comparable with the SOC approaches thus proving clinical feasibility of the proposed new reconstruction Methods: Improved CNR and higher SUVmax in 14 of 18 regions were observed. Different noise texture, as compared to SOC reconstructed images, was observed in low count regions when using OS-SRR-PGA; however, this did not result in false positives or lower lesion conspicuity. Conclusions: This initial study indicates that the application of advanced reconstruction methods may provide equivalent diagnostic and quantitative pediatric PET image quality, without detriment to small lesions detection and quantification, with only one-third the injected [18F]-FDG activity, as compared to SOC approaches. We expect that further refinements of our low-dose OS-SRR-PGA algorithm might allow an even larger reduction of pediatric PET radiation dose, and it is an aim of our ongoing research effort.