Uncertainty Assessment In Whole-Body Low Dose Pet Reconstruction Using Non-Parametric Bayesian Deep Learning Approach

Uncertainty quantification in deep learning (DL) based reconstruction of Positron emission tomography (PET) images plays a critical role in its deployment in clinical practice. Currently, available approaches for uncertainty estimation in DNN-based PET image reconstruction algorithms may result in sub-optimal clinical decision making due to potentially inaccurate estimation of the reconstructed image estimation uncertainty. We introduce a fully non-parametric Bayesian framework for uncertainty estimation in DNN-based Ultra Low Dose PET image reconstruction by combining an Adam optimizer with stochastic gradient Langevin dynamics (SGLD) to characterize the underlying posterior distribution through posterior sampling. The Bayesian LD PET reconstruction approach shows a better correlation of the predicted uncertainty to the Dose Reduction Factor (DRF), as compared to benchmark Monte-Carlo Dropout 95 percentile score (r2=0.9174 vs. r2=0.6144). Furthermore, consistent improvement was achieved in SSIM, PSNR and NRMSE measures for DRF’s 4, 10, 20, 50 and 100.