Bayesian inference for 3D activity reconstruction combining X-ray Computed Tomography and Angular Segmented Gamma Scanning

This study evaluates a Bayesian variational framework for 3D activity reconstruction in 220-liter radioactive waste drums using Angular Segmented Gamma Scanning (ASGS) and attenuation information derived from X-ray Computed Tomography (CT). Computations are made tractable using stochastic variational inference together with a multi-resolution spatial prior for the inferred activities. A CT-to-LAC model is developed to convert CT-derived gray levels into energy-dependent linear attenuation coefficients, allowing the forward model to be constructed under experimental conditions. Tests were performed using homogeneous and heterogeneous mock-ups, with point and linear sources made of either 241Am or 152Eu. To the best of our knowledge, this is the first experimental validation of Bayesian 3D activity reconstruction from coupled ASGS and CT measurements for radiological waste characterization. Results show that the approach recovers spatial activity distributions broadly consistent with known source locations. Regarding total activity, for the multi-line isotope 152Eu, the true activity is recovered within the posterior distribution and the posterior mean outperforms conventional averaging for this multi-line isotope. Posterior mean relative errors ranged from 0.89% to 8.00% across the homogeneous and heterogeneous mock-ups, compared with errors up to 27.19% for the conventional averaged estimate. Posterior standard deviations on total activity were typically about 4%–6% of the posterior mean, indicating well-constrained estimates. In contrast, for 241Am, reconstructed from the single 59.54 keV emission line, posterior mean relative errors ranged from 26.27% to 42.29%, reflecting combined single-line non-identifiability and stronger sensitivity to attenuation-model errors at low energy.