ELECTRIC: Physics-Guided Bayesian CT Reconstruction Reduces Error by 70%

Researchers have introduced ELECTRIC (Evidential Learning-Enhanced CT Reconstruction via Iterative Correction), a novel physics-guided Bayesian formulation for computed tomography (CT) reconstruction. The method employs an evidential neural network to generate an image proposal and an error-predictive epistemic-uncertainty surrogate. This surrogate is transformed into an adaptive precision field, which is integrated into a Poisson-weighted maximum a posteriori (MAP) update. The iterative loop treats prior confidence as a learned state variable, enhancing reconstruction accuracy. In simulation studies using image slices from the AAPM Mayo Clinic Low-Dose CT dataset, the approach demonstrated a mechanism-validation pilot with transparent surrogate estimators and a feasibility study with a trained Normal-Inverse-Gamma evidential network. On held-out patients, the learned prior mean reduced reconstruction error by approximately 70% relative to filtered back-projection. The work includes theoretical analysis and is available on arXiv under the identifier 2608.00060.