PT  - JOURNAL ARTICLE
AU  - Son, Jungah
AU  - Lee, Jae Sung
AU  - Kim, Soo Mee
AU  - Kim, Joong Hyun
AU  - Lee, Dong Soo
TI  - Comparison of gap compensation methods for small-diameter PET
DP  - 2011 May 01
TA  - Journal of Nuclear Medicine
PG  - 1994--1994
VI  - 52
IP  - supplement 1
4099  - http://jnm.snmjournals.org/content/52/supplement_1/1994.short
4100  - http://jnm.snmjournals.org/content/52/supplement_1/1994.full
SO  - J Nucl Med2011 May 01; 52
AB  - 1994 Objectives Gaps between the detector modules in small-ring PET scanners cause significant loss of measured sinogram and lead to significant artifacts in reconstructed images. Several approaches have been suggested to fill in the missing data of the sinogram in Fourier or discrete cosine transform (DCT) domain or to reduce the gap-artifacts during the image reconstruction by incorporating prior information. In this study, we compared the performance of 3 gap compensation methods: (i) DCT gap-filling filter (DGF), (ii) block sequential regularized expectation maximization (BSREM), and (iii) constrained total variation (TV) reconstruction algorithms for the missing data. Methods These 3 methods were applied to the compensation of missing data due to the gaps of 9.2 degree in a MPPC PET system (under development in our group) which consisted of 8 block-detectors with the diameter of 9 cm. For BSREM, Bouman and Sauer prior function was used. TV algorithm was incorporated into ART and EM reconstruction methods. Using Shepp-Logan (SL) phantom and real uniform phantom and brain data (acquired using clinical PET without gap), gap-compensation performance of 3 methods were evaluated in terms of percent error (PE) and uniformity recovery rate. Results TV-ART for noiseless data of SL phantom yielded the lowest PE at same iteration number, however, in case of noisy SL data and real uniform and brain data, TV-ART showed more deteriorated image quality by the quantum noise than other methods. TV-EM and BSREM showed the highest uniformity for uniform phantom and the best qualitative image quality for brain data, respectively. The under-compensation of the missing data by the DGF resulted in artifacts, such as the stripe pattern in reconstructed images. Conclusions All the methods leaded to an improvement of image quality in the noiseless SL data. However, for the noisy SL and real data, TV-ART was most susceptible to the counting statistics. On the contrary, BSREM and TV-EM yielded most reasonable results both in the quantitative and qualitative aspects for the gap compensation of real data