Statistical energy-based scatter correction in 3D list-mode PET

Abstract

100

Objectives: We present a novel fully 3D scatter correction that uses the energy of individual photons detected in list-mode TOF-PET, and compare its accuracy to single scatter simulation (SSS) correction.

Methods: Energy spectra of scattered and primary single photons were estimated using constrained spectral factor analysis (SFA) for different angular bins in the scanner. Next, these spectra were used to compute the probability density functions of primary and scattered photons’ energies. These probabilities were incorporated in two MLEM update equations to jointly reconstruct scatter and primary images. We assessed the accuracy of our method (SFA-MLEM) in Monte Carlo simulations of the TF-Gemini scanner and a cylindrical torso phantom. We compared contrast values obtained with SFA-MLEM to those with SSS, with and without fitting the SSS sinogram to the tails of measurements. We also evaluated the gain in accuracy for estimating contrast when combining SSS with the new energy-based method.

Results: Estimation error on energy spectra was 7% with SFA. Contrast values were 1.89 (ref=2), 3.71 (ref=4), 0.21 (cold) with SFA-MLEM and 1.91, 3.76, 0.25 with SSS/tails fitting. Combining SFA-MLEM and SSS improved the accuracy of contrast estimation from 1.69, 3.63, 0.26 (SSS/no tails fitting) to 1.77, 3.71, 0.24 (SSS/no tails fitting+SFA).

Conclusions: Our results suggest that fully 3D statistical estimation of the scatter and primary images using the energy of detected photons is accurate. This approach accounts for out of field activity and precludes the need to estimate a 3D scatter sinogram to be scaled to the tails of measurements. It also has the potential of correcting for spurious activity when imaging dirty isotopes, such as Y86, by separating the energy spectra of primary and cascade gamma photons.

Research Support: NIH-EB005876, AHA-0655909T