Application of total-variation penalized maximum likelihood image reconstruction for dynamic small animal stationary SPECT studies

Objectives To apply a total-variation (TV) penalized maximum likelihood (ML) image reconstruction method to dynamic small animal (SA) stationary SPECT and evaluate the improvement in pixel-wise time-activity-curve (TAC) estimation compared to conventional ML method.

Methods We implemented an ML+ βTV image reconstruction method whose objective function consisted of the negative Poisson log-likelihood and a β-weighted TV penalty. Our implementation used the Douglas-Rachford splitting method to deal with the non-smooth TV term. We evaluated the ML+ βTV method for dynamic SA stationary SPECT using analytic simulation and the 3D MOBY phantom. The SPECT system geometry modeled a 2nd generation SPECT insert for simultaneous SA SPECT-MR imaging. The analytic simulator included effects of the spatially variant pinhole response and Poisson noise. We simulated a 20-min dynamic mouse renal SPECT scan with 99mTc-DTPA; projection data were collected into 59 equal time frames at 20 sec intervals. The simulated TACs of the mouse’s left and right kidneys were obtained from our previous animal experiment. We reconstructed the 3D volume at individual time frames using the ML+ βTV method and the conventional ML-EM. Pixel-wise time activity curves of the left and the right kidneys were computed. To ensure a fair comparison, the β value in the ML+ βTV method was chosen by minimizing the total squared-error in the regions-of-interest (left/right kidneys). In the ML-EM method, the iteration numbers for the different frames in the TAC computation were determined similarly.

Results The pixel-wise TAC obtained by the ML-EM method exhibited large fluctuation around the truth; this large fluctuation was significantly diminished by using ML+ βTV.

Conclusions We demonstrated significant improvement in pixel-wise TAC estimation by using ML+ βTV method which encourages piece-wise constant image reconstruction. An implied benefit which will be pursued is improvement in subsequent quantitative analysis by combining with kidney kinetic models.

Research Support NIH R01 EB873