RT Journal Article
SR Electronic
T1 Evaulaion of Compressed Sensing Image Acquisition and Processing for Single Photon Emission Tomography
JF Journal of Nuclear Medicine
JO J Nucl Med
FD Society of Nuclear Medicine
SP 534
OP 534
VO 57
IS supplement 2
A1 Song, Bongyong
A1 Ahmad, Ehatsham
A1 Rupe, Eric
A1 Nguyen, Brian
A1 Brunsing, Ryan
A1 Sung, Andrew
A1 Moradi, Farshad
A1 Hoh, Carl
A1 Obrzut, Sebastian
YR 2016
UL http://jnm.snmjournals.org/content/57/supplement_2/534.abstract
AB 534Objectives SPECT images are corrupted by Poisson noise due to inherent inefficiency of the gamma camera, requiring either long imaging times or greater injected radiopharmaceutical doses. The aim of this study is to evaluate compressed sensing image acquisition and processing algorithm for SPECT to reduce image acquisition time. Methods Jaszczak phantom was filled with Tc99m Pertechnetate and saline and imaged with a SPECT camera. SPECT images were reconstructed using ordered subset expectation maximization (OSEM), filtered back projection (FBP), Simultaneous Algebraic Reconstruction Technique (SART) and compressed sensing (CS) with Total Variation (TV) minimization algorithm utilizing 128, 64 and 32 projections. Reconstructed images were ranked by 7 observers based on visual quality and mean ranking scores were compared. Furthermore, virtual Jaszczak phantom was used to model varying levels of Poisson noise based on image acquisition times. Root mean squared error (RMSE) was calculated and compared for OSEM, FBP, SART and CS reconstructions of virtual phantom with various numbers of projections. Results 128 projection CS reconstruction of real Jaszczak phantom is visually superior to 128 projection OSEM (mean large rod ranking score (LRRS) 1.86+/-0.90 vs. 6.00+/-1.53 p=0.0001 and mean small rod ranking score (SRRS) 1.86+/-0.69 vs. 3.57+/-1.81 p=0.0377). 64 projection CS reconstruction of real Jaszczak phantom is visually equivalent to 128 projection OSEM (LRRS 5.14+/-2.54 vs. 6.00+/-1.53 p=0.4577 and SRRS 4.43+/-0.79 vs. 3.57+/-1.81 p=0.2717). RMSE of 0.1211 for 64 projection CS reconstruction was smaller than RMSE of 0.1236 for 128 projection OSEM of large rod virtual phantom with Poisson noise model. RMSE of 0.1192 for 64 projection CS reconstruction was smaller than RMSE of 0.1318 for 128 projection OSEM of small rod virtual phantom with Poisson noise model. At image acquisition time reduced in half and increased Poisson noise, RMSE of 0.1322 for 64 projection CS reconstruction was smaller than RMSE of 0.1324 for 128 projection OSEM of small rod virtual phantom. Conclusions Compressed sensing reconstruction of SPECT images utilizing reduced number of projections could potentially be performed without loss of image quality compared to standard acquisition and OSEM reconstruction, resulting in decreased overall image acquisition time. The CS reconstruction is applicable to existing scanners without requiring hardware changes.