Total variation penalized maximum-likelihood image reconstruction for a stationary small animal SPECT system

We have developed a 3D total variation (TV) penalized maximum likelihood (ML) image reconstruction method and tested it in simulated dynamic SPECT scans using a stationary ring-type SPECT insert for simultaneous small animal SPECTMR imaging. The SPECT insert consists of 5 (axial) x 19 (transaxial) MR-compatible CZT detectors that form a seamless 19-side detector ring, inside which a cylindrical collimator sleeve with 36 focused pinholes is inserted for dynamic SPECT acquisition. The short duration of the individual time frames and the stationary data acquisition nature may cause severe noise and sparse-view artifacts in the reconstructed images, and as a result affect the time-activity curve (TAC) derived from the 4D image sequence. The TV potential function favors piece-wise constant image reconstruction therefore is capable of reducing these image artifacts. Our implementation of the ML-TV method used the Douglas-Rachford splitting to deal with the non-smooth TV function. We applied the ML-TV method to a computer-simulated dynamic mouse renal SPECT scan, and evaluated the method in terms of pixel-wise TAC estimation compared to the conventional ML-EM. The pixel-wise TAC obtained by the MLEM method exhibited large fluctuation around the truth; this large fluctuation was significantly suppressed by using ML-TV. Our next step is to incorporate time-domain correlation and develop fully 4D (3D spatial + 1D time) image reconstruction methods for dynamic stationary small animal SPECT studies. (orig.)