Evaluation of a maximum a posteriori (MAP) reconstruction method using an anatomical prior for improved dose-rate volume histogram estimation

Objectives Advanced image-based radiobiological dosimetry promises improved prediction of normal organ response in targeted radionuclide therapy (TRT), but requires accurate estimates of the activity distribution as an input. ECT images are, however, degraded by partial volume effects (PVEs) and noise. Availability of multimodality ECT/CT systems provides high resolution anatomical information that can potentially be used to improve activity estimates. In previous work, we proposed a convergent accelerated MAP algorithm for a region mean prior to improve activity distribution estimates. The prior penalizes deviations from the mean in a volume of interest (VOI). Here we evaluated the algorithm in the context of TRT for the case of non-uniform activity distributions inside VOIs.

Methods We investigated 2 types of non-uniform activity distributions: liver uptake with a lumpy texture and a tumor with a necrotic core. We used the 3D XCAT phantom to model patient anatomy. We generated simulated analytic In-111 projection data that modeled image degrading effects in SPECT. Images were reconstructed with compensation for degrading effects using both the proposed MAP method and OS-EM. OS-EM images were optionally post-filtered. Dose-rate distributions were estimated and methods were compared in terms of the shape and fidelity of the cumulative dose-rate volume histogram for the liver and tumor using optimal filtering and prior weight parameters.

Results The proposed MAP algorithm substantially reduced image noise and PVEs and thus more faithfully reproduced the histograms, even with non-uniform activity distributions inside VOIs. The MAP method reduced PVEs when boundaries were known and exhibited PVEs comparable to OS-EM when boundaries were not known.

Conclusions The MAP reconstruction algorithm based on a region mean prior provided improved dose rate estimates compared to OS-EM even for VOIs with non-uniform activity distributions