Partial volume correction in small animal PET imaging incorporating total variation regularization

Objectives Partial volume correction (PVC) methods in emission tomography aim to achieve enhanced resolution and contrast recovery. However, traditional iterative deconvolution algorithms such as Van Cittert (VC) and Richardson-Lucy (RL) commonly amplify noise. The aim of this study was to incorporate total variation (TV) regularization within an iterative deconvolution framework and compare performance with a range of other PVC methods.

Methods In order to gain control over noise levels, we incorporated TV regularization within VC and RL deconvolution algorithms, and developed 3D VC-TV and RL-TV deconvolution. For validation, a 20 minutes NEMA NU 4-2008 IQ phantom with 18FDG as well as 10 minute tumor mouse data were acquired via the Siemens Inveon micro PET and were subsequently reconstructed using the OSEM algorithm. First, we optimized the full width at half maximum (FWHM) for RL and VC deconvolution. Then, we optimized the performance of the proposed TV regularized deconvolution (both VC-TV and RL-TV) and compared it with median root prior (MRP) based RL algorithm and El Naqa’s Bayesian deconvolution (Ba) algorithm. Comparisons included both visual assessment and recovery coefficient (RC) vs. standard deviation (SD) increase trade-off performance.

Results The experimental results demonstrated the proposed algorithms (both VC-TV and RL-TV)to achieve superior visual and quantitative performancecompared with the MRP-based RL and Bayesian deconvolution algorithms. In particular, the RL-TVexhibitedthe best compromise between noise attenuation and intensity recovery forthe images.

Conclusions The incorporation of total variation regularization in RL deconvolution algorithm is an efficient PVC method in 3D small animal PET imaging.

Research Support This work was supported by the 973 Program of China under grant no. 2010CB732504.